transform learning mri with global wavelet regularization
play

Transform Learning MRI with Global Wavelet Regularization A. Korhan - PowerPoint PPT Presentation

Aug 02, 2023 •191 likes •1.58k views

Introduction Transform Learning MRI GTLMRI Simulations and Conclusion Transform Learning MRI with Global Wavelet Regularization A. Korhan Tanc 1 Ender M. Eksioglu 2 1 Department of EEE Kirklareli University Kirklareli, Turkey 2 Department of

  1. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Sparse MRI 2 � F u x − y � 2 1 min 2 + ρ 1 � Φ x � 1 + ρ 2 � x � TV . x x ∈ C N is the reconstructed MR image in vectorized form. F u is the undersampled Fourier transform operator: conversion from the vectorized image to the k-space. y = F u x ⋆ + η ∈ C κ is the observation vector in the k-space. x ⋆ is the true underlying image and η is the additive noise. The ratio κ/ N quantifies the undersampling. �·� 1 denotes the ℓ 1 norm. Φ is a sparsifying operator: we will assume it to be a square wavelet transform. �·� TV is the Total Variation (TV) norm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  2. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Sparse MRI 2 � F u x − y � 2 1 min 2 + ρ 1 � Φ x � 1 + ρ 2 � x � TV . x x ∈ C N is the reconstructed MR image in vectorized form. F u is the undersampled Fourier transform operator: conversion from the vectorized image to the k-space. y = F u x ⋆ + η ∈ C κ is the observation vector in the k-space. x ⋆ is the true underlying image and η is the additive noise. The ratio κ/ N quantifies the undersampling. �·� 1 denotes the ℓ 1 norm. Φ is a sparsifying operator: we will assume it to be a square wavelet transform. �·� TV is the Total Variation (TV) norm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  3. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Sparse MRI 2 � F u x − y � 2 1 min 2 + ρ 1 � Φ x � 1 + ρ 2 � x � TV . x x ∈ C N is the reconstructed MR image in vectorized form. F u is the undersampled Fourier transform operator: conversion from the vectorized image to the k-space. y = F u x ⋆ + η ∈ C κ is the observation vector in the k-space. x ⋆ is the true underlying image and η is the additive noise. The ratio κ/ N quantifies the undersampling. �·� 1 denotes the ℓ 1 norm. Φ is a sparsifying operator: we will assume it to be a square wavelet transform. �·� TV is the Total Variation (TV) norm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  4. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Outline Introduction 1 The Problem The Novel Approach Transform Learning MRI 2 GTLMRI 3 New Cost GTLMRI: Denoising GTLMRI: Reconstruction GTLMRI: Overall Algorithm Simulations and Conclusion 4 Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  5. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Patch based regularization methods Examplar or patch based methods have been very popular for sparsity based image processing. Dictionary learning (DL) based synthesis sparsity methods Analysis sparsity based analysis operator learning methods Novel model for analysis operator learning, called as sparsifying Transform Learning (TL) [Ravishankar and Bresler, 2013]. TL has been utilized to regularize the MRI reconstruction problem, resulting in the TLMRI algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  6. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Patch based regularization methods Examplar or patch based methods have been very popular for sparsity based image processing. Dictionary learning (DL) based synthesis sparsity methods Analysis sparsity based analysis operator learning methods Novel model for analysis operator learning, called as sparsifying Transform Learning (TL) [Ravishankar and Bresler, 2013]. TL has been utilized to regularize the MRI reconstruction problem, resulting in the TLMRI algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  7. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Patch based regularization methods Examplar or patch based methods have been very popular for sparsity based image processing. Dictionary learning (DL) based synthesis sparsity methods Analysis sparsity based analysis operator learning methods Novel model for analysis operator learning, called as sparsifying Transform Learning (TL) [Ravishankar and Bresler, 2013]. TL has been utilized to regularize the MRI reconstruction problem, resulting in the TLMRI algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  8. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Patch based regularization methods Examplar or patch based methods have been very popular for sparsity based image processing. Dictionary learning (DL) based synthesis sparsity methods Analysis sparsity based analysis operator learning methods Novel model for analysis operator learning, called as sparsifying Transform Learning (TL) [Ravishankar and Bresler, 2013]. TL has been utilized to regularize the MRI reconstruction problem, resulting in the TLMRI algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  9. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Patch based regularization methods Examplar or patch based methods have been very popular for sparsity based image processing. Dictionary learning (DL) based synthesis sparsity methods Analysis sparsity based analysis operator learning methods Novel model for analysis operator learning, called as sparsifying Transform Learning (TL) [Ravishankar and Bresler, 2013]. TL has been utilized to regularize the MRI reconstruction problem, resulting in the TLMRI algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  10. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Patch based regularization methods Methods such as Sparse MRI, RecPF and FCSA apply global, image-scale regularization TLMRI or DL based algorithms utilize local, patch-scale regularization In this work, we aim to bring these two ends together. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  11. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Patch based regularization methods Methods such as Sparse MRI, RecPF and FCSA apply global, image-scale regularization TLMRI or DL based algorithms utilize local, patch-scale regularization In this work, we aim to bring these two ends together. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  12. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion Patch based regularization methods Methods such as Sparse MRI, RecPF and FCSA apply global, image-scale regularization TLMRI or DL based algorithms utilize local, patch-scale regularization In this work, we aim to bring these two ends together. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  13. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion New Method: Globally regularized TLMRI G-TLMRI We introduce a global sparsifying cost into TLMRI, and provide the algorithm. We will denote this modified framework as the Globally regularized TLMRI (G-TLMRI). Simulation results: use of global and local regularization terms together results in superior reconstruction performance. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  14. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion New Method: Globally regularized TLMRI G-TLMRI We introduce a global sparsifying cost into TLMRI, and provide the algorithm. We will denote this modified framework as the Globally regularized TLMRI (G-TLMRI). Simulation results: use of global and local regularization terms together results in superior reconstruction performance. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  15. Introduction Transform Learning MRI The Problem GTLMRI The Novel Approach Simulations and Conclusion New Method: Globally regularized TLMRI G-TLMRI We introduce a global sparsifying cost into TLMRI, and provide the algorithm. We will denote this modified framework as the Globally regularized TLMRI (G-TLMRI). Simulation results: use of global and local regularization terms together results in superior reconstruction performance. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  16. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI TL has been applied to MRI image reconstruction. TLMRI cost function can be stated as follows. � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . (2) Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  17. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI TL has been applied to MRI image reconstruction. TLMRI cost function can be stated as follows. � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . (2) Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  18. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI TL has been applied to MRI image reconstruction. TLMRI cost function can be stated as follows. � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . (2) Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  19. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI TL has been applied to MRI image reconstruction. TLMRI cost function can be stated as follows. � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . (2) Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  20. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . �·� F is the Frobenius matrix norm. �·� 0 denotes the ℓ 0 pseudo-norm. W ∈ C n × n is the learned square transform. x j ∈ C n denote vectorized 2D X ∈ C n × M , and its columns ˆ ˆ patches of size √ n × √ n . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  21. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . �·� F is the Frobenius matrix norm. �·� 0 denotes the ℓ 0 pseudo-norm. W ∈ C n × n is the learned square transform. x j ∈ C n denote vectorized 2D X ∈ C n × M , and its columns ˆ ˆ patches of size √ n × √ n . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  22. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . �·� F is the Frobenius matrix norm. �·� 0 denotes the ℓ 0 pseudo-norm. W ∈ C n × n is the learned square transform. x j ∈ C n denote vectorized 2D X ∈ C n × M , and its columns ˆ ˆ patches of size √ n × √ n . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  23. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . �·� F is the Frobenius matrix norm. �·� 0 denotes the ℓ 0 pseudo-norm. W ∈ C n × n is the learned square transform. x j ∈ C n denote vectorized 2D X ∈ C n × M , and its columns ˆ ˆ patches of size √ n × √ n . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  24. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . �·� F is the Frobenius matrix norm. �·� 0 denotes the ℓ 0 pseudo-norm. W ∈ C n × n is the learned square transform. x j ∈ C n denote vectorized 2D X ∈ C n × M , and its columns ˆ ˆ patches of size √ n × √ n . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  25. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . �·� F is the Frobenius matrix norm. �·� 0 denotes the ℓ 0 pseudo-norm. W ∈ C n × n is the learned square transform. x j ∈ C n denote vectorized 2D X ∈ C n × M , and its columns ˆ ˆ patches of size √ n × √ n . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  26. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . A ∈ C n × M includes the sparse codes. Q ( · ) penalization term for the learned W . R image to patch operator. Observation fidelity is enforced using the � F u x − y � 2 2 term. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  27. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . A ∈ C n × M includes the sparse codes. Q ( · ) penalization term for the learned W . R image to patch operator. Observation fidelity is enforced using the � F u x − y � 2 2 term. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  28. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . A ∈ C n × M includes the sparse codes. Q ( · ) penalization term for the learned W . R image to patch operator. Observation fidelity is enforced using the � F u x − y � 2 2 term. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  29. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . A ∈ C n × M includes the sparse codes. Q ( · ) penalization term for the learned W . R image to patch operator. Observation fidelity is enforced using the � F u x − y � 2 2 term. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  30. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . A ∈ C n × M includes the sparse codes. Q ( · ) penalization term for the learned W . R image to patch operator. Observation fidelity is enforced using the � F u x − y � 2 2 term. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  31. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI � W ˆ X − A � 2 F + λ Q ( W ) + τ � R ( x ) − ˆ X � 2 (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . A ∈ C n × M includes the sparse codes. Q ( · ) penalization term for the learned W . R image to patch operator. Observation fidelity is enforced using the � F u x − y � 2 2 term. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  32. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI TLMRI applies local regularization via a learned sparsifying transform. TLMRI with learned, local regularization: good performance when compared to nonadaptive global regularization (such as wavelet plus TV regularization in Sparse MRI). In this work: include additional global regularization in the TLMRI framework. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  33. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI TLMRI applies local regularization via a learned sparsifying transform. TLMRI with learned, local regularization: good performance when compared to nonadaptive global regularization (such as wavelet plus TV regularization in Sparse MRI). In this work: include additional global regularization in the TLMRI framework. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  34. Introduction Transform Learning MRI GTLMRI Simulations and Conclusion From the Literature: Transform Learning MRI TLMRI TLMRI applies local regularization via a learned sparsifying transform. TLMRI with learned, local regularization: good performance when compared to nonadaptive global regularization (such as wavelet plus TV regularization in Sparse MRI). In this work: include additional global regularization in the TLMRI framework. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  35. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm Outline Introduction 1 The Problem The Novel Approach Transform Learning MRI 2 GTLMRI 3 New Cost GTLMRI: Denoising GTLMRI: Reconstruction GTLMRI: Overall Algorithm Simulations and Conclusion 4 Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  36. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI New cost function with global regularizer. X − A � 2 � W ˆ (P1) min F + λ Q ( W ) + β � A � 1 W , ˆ X , A , x X � 2 F + η � F u x − y � 2 + τ � R ( x ) − ˆ 2 + υ ′ � Φ x � 1 . (3) Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  37. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI New cost function with global regularizer. X − A � 2 � W ˆ (P1) min F + λ Q ( W ) + β � A � 1 W , ˆ X , A , x X � 2 F + η � F u x − y � 2 + τ � R ( x ) − ˆ 2 + υ ′ � Φ x � 1 . (3) Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  38. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI New cost function with global regularizer. X − A � 2 � W ˆ (P1) min F + λ Q ( W ) + β � A � 1 W , ˆ X , A , x X � 2 F + η � F u x − y � 2 + τ � R ( x ) − ˆ 2 + υ ′ � Φ x � 1 . (3) Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  39. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI X − A � 2 � W ˆ (P1) min F + λ Q ( W ) + β � A � 1 W , ˆ X , A , x X � 2 F + η � F u x − y � 2 + τ � R ( x ) − ˆ 2 + υ ′ � Φ x � 1 . X − A � 2 X � 2 � W ˆ F + λ Q ( W ) + τ � R ( x ) − ˆ (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . When compared with (P0), in (P1) the crucial change is the introduction of the � Φ x � 1 term. We will denote this modified framework as the Globally regularized TLMRI (G-TLMRI). Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  40. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI X − A � 2 � W ˆ (P1) min F + λ Q ( W ) + β � A � 1 W , ˆ X , A , x X � 2 F + η � F u x − y � 2 + τ � R ( x ) − ˆ 2 + υ ′ � Φ x � 1 . X − A � 2 X � 2 � W ˆ F + λ Q ( W ) + τ � R ( x ) − ˆ (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . When compared with (P0), in (P1) the crucial change is the introduction of the � Φ x � 1 term. We will denote this modified framework as the Globally regularized TLMRI (G-TLMRI). Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  41. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI X − A � 2 � W ˆ (P1) min F + λ Q ( W ) + β � A � 1 W , ˆ X , A , x X � 2 F + η � F u x − y � 2 + τ � R ( x ) − ˆ 2 + υ ′ � Φ x � 1 . X − A � 2 X � 2 � W ˆ F + λ Q ( W ) + τ � R ( x ) − ˆ (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . When compared with (P0), in (P1) the crucial change is the introduction of the � Φ x � 1 term. We will denote this modified framework as the Globally regularized TLMRI (G-TLMRI). Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  42. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI X − A � 2 � W ˆ (P1) min F + λ Q ( W ) + β � A � 1 W , ˆ X , A , x X � 2 F + η � F u x − y � 2 + τ � R ( x ) − ˆ 2 + υ ′ � Φ x � 1 . X − A � 2 X � 2 � W ˆ F + λ Q ( W ) + τ � R ( x ) − ˆ (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . When compared with (P0), in (P1) the crucial change is the introduction of the � Φ x � 1 term. We will denote this modified framework as the Globally regularized TLMRI (G-TLMRI). Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  43. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI X − A � 2 � W ˆ (P1) min F + λ Q ( W ) + β � A � 1 W , ˆ X , A , x X � 2 F + η � F u x − y � 2 + τ � R ( x ) − ˆ 2 + υ ′ � Φ x � 1 . X − A � 2 X � 2 � W ˆ F + λ Q ( W ) + τ � R ( x ) − ˆ (P0) min F W , ˆ X , A , x + η � F u x − y � 2 2 , s . t . � α j � 0 ≤ s j ∀ j = 1 . . . M . When compared with (P0), in (P1) the crucial change is the introduction of the � Φ x � 1 term. We will denote this modified framework as the Globally regularized TLMRI (G-TLMRI). Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  44. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI We will separate the algorithm into two steps with and without optimization on x . X − A � 2 X � 2 � W ˆ F + λ Q ( W )+ β � A � 1 + τ � R ( x ) − ˆ (P2) min F . (4) W , ˆ X , A 1 2 � F u x − y � 2 2 η � R ( x ) − ˆ X � 2 F + υ ′ 2 + τ 2 η � Φ x � 1 . (5) (P3) min x (P2) can be thought of as denoising. (P3) can be thought of as reconstruction. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  45. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI We will separate the algorithm into two steps with and without optimization on x . X − A � 2 X � 2 � W ˆ F + λ Q ( W )+ β � A � 1 + τ � R ( x ) − ˆ (P2) min F . (4) W , ˆ X , A 1 2 � F u x − y � 2 2 η � R ( x ) − ˆ X � 2 F + υ ′ 2 + τ 2 η � Φ x � 1 . (5) (P3) min x (P2) can be thought of as denoising. (P3) can be thought of as reconstruction. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  46. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI We will separate the algorithm into two steps with and without optimization on x . X − A � 2 X � 2 � W ˆ F + λ Q ( W )+ β � A � 1 + τ � R ( x ) − ˆ (P2) min F . (4) W , ˆ X , A 1 2 � F u x − y � 2 2 η � R ( x ) − ˆ X � 2 F + υ ′ 2 + τ 2 η � Φ x � 1 . (5) (P3) min x (P2) can be thought of as denoising. (P3) can be thought of as reconstruction. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  47. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI We will separate the algorithm into two steps with and without optimization on x . X − A � 2 X � 2 � W ˆ F + λ Q ( W )+ β � A � 1 + τ � R ( x ) − ˆ (P2) min F . (4) W , ˆ X , A 1 2 � F u x − y � 2 2 η � R ( x ) − ˆ X � 2 F + υ ′ 2 + τ 2 η � Φ x � 1 . (5) (P3) min x (P2) can be thought of as denoising. (P3) can be thought of as reconstruction. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  48. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI We will separate the algorithm into two steps with and without optimization on x . X − A � 2 X � 2 � W ˆ F + λ Q ( W )+ β � A � 1 + τ � R ( x ) − ˆ (P2) min F . (4) W , ˆ X , A 1 2 � F u x − y � 2 2 η � R ( x ) − ˆ X � 2 F + υ ′ 2 + τ 2 η � Φ x � 1 . (5) (P3) min x (P2) can be thought of as denoising. (P3) can be thought of as reconstruction. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  49. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm New Method: GTLMRI We will separate the algorithm into two steps with and without optimization on x . X − A � 2 X � 2 � W ˆ F + λ Q ( W )+ β � A � 1 + τ � R ( x ) − ˆ (P2) min F . (4) W , ˆ X , A 1 2 � F u x − y � 2 2 η � R ( x ) − ˆ X � 2 F + υ ′ 2 + τ 2 η � Φ x � 1 . (5) (P3) min x (P2) can be thought of as denoising. (P3) can be thought of as reconstruction. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  50. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm Outline Introduction 1 The Problem The Novel Approach Transform Learning MRI 2 GTLMRI 3 New Cost GTLMRI: Denoising GTLMRI: Reconstruction GTLMRI: Overall Algorithm Simulations and Conclusion 4 Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  51. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising We will divide (P2) into two in the following form similar to the TLMRI. W , A � W ˆ X − A � 2 F + λ Q ( W ) + β � A � 1 . (P2.1) min � W ˆ X − A � 2 F + β � A � 1 + τ � R ( x ) − ˆ X � 2 F . (P2.2) min ˆ X , A Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  52. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising We will divide (P2) into two in the following form similar to the TLMRI. W , A � W ˆ X − A � 2 F + λ Q ( W ) + β � A � 1 . (P2.1) min � W ˆ X − A � 2 F + β � A � 1 + τ � R ( x ) − ˆ X � 2 F . (P2.2) min ˆ X , A Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  53. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising We will divide (P2) into two in the following form similar to the TLMRI. W , A � W ˆ X − A � 2 F + λ Q ( W ) + β � A � 1 . (P2.1) min � W ˆ X − A � 2 F + β � A � 1 + τ � R ( x ) − ˆ X � 2 F . (P2.2) min ˆ X , A Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  54. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising We will divide (P2) into two in the following form similar to the TLMRI. W , A � W ˆ X − A � 2 F + λ Q ( W ) + β � A � 1 . (P2.1) min � W ˆ X − A � 2 F + β � A � 1 + τ � R ( x ) − ˆ X � 2 F . (P2.2) min ˆ X , A Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  55. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising We will divide (P2) into two in the following form similar to the TLMRI. W , A � W ˆ X − A � 2 F + λ Q ( W ) + β � A � 1 . (P2.1) min � W ˆ X − A � 2 F + β � A � 1 + τ � R ( x ) − ˆ X � 2 F . (P2.2) min ˆ X , A Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  56. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising (P2.1) can be approximately solved using iterative alternation over two steps. X − A � 2 A � W ˆ (P2.1.1) min F + β � A � 1 . W � W ˆ X − A � 2 F + λ Q ( W ) . (P2.1.2) min Both (P2.1.1) and (P2.1.2) have closed form solutions. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  57. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising (P2.1) can be approximately solved using iterative alternation over two steps. X − A � 2 A � W ˆ (P2.1.1) min F + β � A � 1 . W � W ˆ X − A � 2 F + λ Q ( W ) . (P2.1.2) min Both (P2.1.1) and (P2.1.2) have closed form solutions. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  58. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising (P2.1) can be approximately solved using iterative alternation over two steps. X − A � 2 A � W ˆ (P2.1.1) min F + β � A � 1 . W � W ˆ X − A � 2 F + λ Q ( W ) . (P2.1.2) min Both (P2.1.1) and (P2.1.2) have closed form solutions. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  59. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising (P2.1) can be approximately solved using iterative alternation over two steps. X − A � 2 A � W ˆ (P2.1.1) min F + β � A � 1 . W � W ˆ X − A � 2 F + λ Q ( W ) . (P2.1.2) min Both (P2.1.1) and (P2.1.2) have closed form solutions. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  60. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising (P2.1) can be approximately solved using iterative alternation over two steps. X − A � 2 A � W ˆ (P2.1.1) min F + β � A � 1 . W � W ˆ X − A � 2 F + λ Q ( W ) . (P2.1.2) min Both (P2.1.1) and (P2.1.2) have closed form solutions. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  61. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising Two alternating steps for (P2.2) become as follows. A � W ˆ X − A � 2 F + β � A � 1 . (P2.2.1) min � W ˆ X − A � 2 F + τ � R ( x ) − ˆ X � 2 F . (P2.2.2) min ˆ X (P2.2.1) is again solved by soft thresholding. (P2.2.2) has a simple least squares solution for fixed A given by ( W H W + τ I ) − 1 ( W H A + τ R ( x )) . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  62. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising Two alternating steps for (P2.2) become as follows. A � W ˆ X − A � 2 F + β � A � 1 . (P2.2.1) min � W ˆ X − A � 2 F + τ � R ( x ) − ˆ X � 2 F . (P2.2.2) min ˆ X (P2.2.1) is again solved by soft thresholding. (P2.2.2) has a simple least squares solution for fixed A given by ( W H W + τ I ) − 1 ( W H A + τ R ( x )) . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  63. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising Two alternating steps for (P2.2) become as follows. A � W ˆ X − A � 2 F + β � A � 1 . (P2.2.1) min � W ˆ X − A � 2 F + τ � R ( x ) − ˆ X � 2 F . (P2.2.2) min ˆ X (P2.2.1) is again solved by soft thresholding. (P2.2.2) has a simple least squares solution for fixed A given by ( W H W + τ I ) − 1 ( W H A + τ R ( x )) . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  64. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising Two alternating steps for (P2.2) become as follows. A � W ˆ X − A � 2 F + β � A � 1 . (P2.2.1) min � W ˆ X − A � 2 F + τ � R ( x ) − ˆ X � 2 F . (P2.2.2) min ˆ X (P2.2.1) is again solved by soft thresholding. (P2.2.2) has a simple least squares solution for fixed A given by ( W H W + τ I ) − 1 ( W H A + τ R ( x )) . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  65. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising Two alternating steps for (P2.2) become as follows. A � W ˆ X − A � 2 F + β � A � 1 . (P2.2.1) min � W ˆ X − A � 2 F + τ � R ( x ) − ˆ X � 2 F . (P2.2.2) min ˆ X (P2.2.1) is again solved by soft thresholding. (P2.2.2) has a simple least squares solution for fixed A given by ( W H W + τ I ) − 1 ( W H A + τ R ( x )) . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  66. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Denoising Two alternating steps for (P2.2) become as follows. A � W ˆ X − A � 2 F + β � A � 1 . (P2.2.1) min � W ˆ X − A � 2 F + τ � R ( x ) − ˆ X � 2 F . (P2.2.2) min ˆ X (P2.2.1) is again solved by soft thresholding. (P2.2.2) has a simple least squares solution for fixed A given by ( W H W + τ I ) − 1 ( W H A + τ R ( x )) . Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  67. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm Outline Introduction 1 The Problem The Novel Approach Transform Learning MRI 2 GTLMRI 3 New Cost GTLMRI: Denoising GTLMRI: Reconstruction GTLMRI: Overall Algorithm Simulations and Conclusion 4 Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  68. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction The second main step for the solution of (P1) is the reconstruction step, (P3). 2 � F u x − y � 2 X � 2 F + υ ′ 1 2 η � R ( x ) − ˆ (P3) min 2 + τ 2 η � Φ x � 1 . x Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  69. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction The second main step for the solution of (P1) is the reconstruction step, (P3). 2 � F u x − y � 2 X � 2 F + υ ′ 1 2 η � R ( x ) − ˆ (P3) min 2 + τ 2 η � Φ x � 1 . x Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  70. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction The second main step for the solution of (P1) is the reconstruction step, (P3). 2 � F u x − y � 2 X � 2 F + υ ′ 1 2 η � R ( x ) − ˆ (P3) min 2 + τ 2 η � Φ x � 1 . x Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  71. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define patch to image operator ˆ R . j R T R ( ˆ ˆ �� j ˆ � X ) = x j ./ w . (P3) can be approximately rewritten as follows. 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 (P 3 ′ ) min � � + υ � Φ x � 1 . (6) 2 2 x 1 2 � F u x − y � 2 X � 2 F + υ ′ 2 η � R ( x ) − ˆ 2 + τ 2 η � Φ x � 1 . (P3) min x Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  72. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define patch to image operator ˆ R . j R T R ( ˆ ˆ �� j ˆ � X ) = x j ./ w . (P3) can be approximately rewritten as follows. 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 (P 3 ′ ) min � � + υ � Φ x � 1 . (6) 2 2 x 1 2 � F u x − y � 2 X � 2 F + υ ′ 2 η � R ( x ) − ˆ 2 + τ 2 η � Φ x � 1 . (P3) min x Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  73. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define patch to image operator ˆ R . j R T R ( ˆ ˆ �� j ˆ � X ) = x j ./ w . (P3) can be approximately rewritten as follows. 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 (P 3 ′ ) min � � + υ � Φ x � 1 . (6) 2 2 x 1 2 � F u x − y � 2 X � 2 F + υ ′ 2 η � R ( x ) − ˆ 2 + τ 2 η � Φ x � 1 . (P3) min x Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  74. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define patch to image operator ˆ R . j R T R ( ˆ ˆ �� j ˆ � X ) = x j ./ w . (P3) can be approximately rewritten as follows. 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 (P 3 ′ ) min � � + υ � Φ x � 1 . (6) 2 2 x 1 2 � F u x − y � 2 X � 2 F + υ ′ 2 η � R ( x ) − ˆ 2 + τ 2 η � Φ x � 1 . (P3) min x Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  75. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define patch to image operator ˆ R . j R T R ( ˆ ˆ �� j ˆ � X ) = x j ./ w . (P3) can be approximately rewritten as follows. 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 (P 3 ′ ) min � � + υ � Φ x � 1 . (6) 2 2 x 1 2 � F u x − y � 2 X � 2 F + υ ′ 2 η � R ( x ) − ˆ 2 + τ 2 η � Φ x � 1 . (P3) min x Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  76. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define patch to image operator ˆ R . j R T R ( ˆ ˆ �� j ˆ � X ) = x j ./ w . (P3) can be approximately rewritten as follows. 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 (P 3 ′ ) min � � + υ � Φ x � 1 . (6) 2 2 x 1 2 � F u x − y � 2 X � 2 F + υ ′ 2 η � R ( x ) − ˆ 2 + τ 2 η � Φ x � 1 . (P3) min x Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  77. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define two functions: g ( x ) = 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 � � 2 2 f ( x ) = υ � Φ x � 1 . (P 3 ′ ) min f ( x ) + g ( x ) . x This problem can be solved very efficiently by proximal splitting methods. We have used the forward-backward splitting algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  78. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define two functions: g ( x ) = 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 � � 2 2 f ( x ) = υ � Φ x � 1 . (P 3 ′ ) min f ( x ) + g ( x ) . x This problem can be solved very efficiently by proximal splitting methods. We have used the forward-backward splitting algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  79. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define two functions: g ( x ) = 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 � � 2 2 f ( x ) = υ � Φ x � 1 . (P 3 ′ ) min f ( x ) + g ( x ) . x This problem can be solved very efficiently by proximal splitting methods. We have used the forward-backward splitting algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  80. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define two functions: g ( x ) = 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 � � 2 2 f ( x ) = υ � Φ x � 1 . (P 3 ′ ) min f ( x ) + g ( x ) . x This problem can be solved very efficiently by proximal splitting methods. We have used the forward-backward splitting algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  81. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define two functions: g ( x ) = 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 � � 2 2 f ( x ) = υ � Φ x � 1 . (P 3 ′ ) min f ( x ) + g ( x ) . x This problem can be solved very efficiently by proximal splitting methods. We have used the forward-backward splitting algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  82. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define two functions: g ( x ) = 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 � � 2 2 f ( x ) = υ � Φ x � 1 . (P 3 ′ ) min f ( x ) + g ( x ) . x This problem can be solved very efficiently by proximal splitting methods. We have used the forward-backward splitting algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  83. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction Define two functions: g ( x ) = 1 � F u x − y � 2 2 + τ ′ � x − ˆ R ( ˆ X ) � 2 � � 2 2 f ( x ) = υ � Φ x � 1 . (P 3 ′ ) min f ( x ) + g ( x ) . x This problem can be solved very efficiently by proximal splitting methods. We have used the forward-backward splitting algorithm. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  84. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction The forward-backward splitting steps: z = x − γ ∇ g ( x ) . (7) (P3.1) (P3.2) x = x + µ ( prox γ f ( z ) − x ) . (8) ∇ g ( x ) = F H u ( F u x − y )+ τ ′ ( x − ˆ R ( ˆ X )) . F H u is the adjoint operator of F u , it realizes zero-filled reconstruction. prox γ f ( · ) is realized by soft thresholding in the transform ( Φ ) domain and taking an inverse transform. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  85. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction The forward-backward splitting steps: z = x − γ ∇ g ( x ) . (7) (P3.1) (P3.2) x = x + µ ( prox γ f ( z ) − x ) . (8) ∇ g ( x ) = F H u ( F u x − y )+ τ ′ ( x − ˆ R ( ˆ X )) . F H u is the adjoint operator of F u , it realizes zero-filled reconstruction. prox γ f ( · ) is realized by soft thresholding in the transform ( Φ ) domain and taking an inverse transform. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

  86. Introduction New Cost Transform Learning MRI GTLMRI: Denoising GTLMRI GTLMRI: Reconstruction Simulations and Conclusion GTLMRI: Overall Algorithm GTLMRI: Reconstruction The forward-backward splitting steps: z = x − γ ∇ g ( x ) . (7) (P3.1) (P3.2) x = x + µ ( prox γ f ( z ) − x ) . (8) ∇ g ( x ) = F H u ( F u x − y )+ τ ′ ( x − ˆ R ( ˆ X )) . F H u is the adjoint operator of F u , it realizes zero-filled reconstruction. prox γ f ( · ) is realized by soft thresholding in the transform ( Φ ) domain and taking an inverse transform. Tanc and Eksioglu - EUSIPCO 2015 Transform Learning MRI with Global Wavelet Regularization

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Optimizing Discrete Wavelet Transform Optimizing Discrete Wavelet Transform on the Cell Broadband

Optimizing Discrete Wavelet Transform Optimizing Discrete Wavelet Transform on the Cell Broadband

Optimizing Discrete Wavelet Transform Optimizing Discrete Wavelet Transform on the Cell Broadband Engine on the Cell Broadband Engine Seunghwa Kang David A. Bader Key Contributions We design an efficient data decomposition scheme

586 views • 27 slides
GPU-Accelerated Undecimated Wavelet Transform for Film and Video Denoising Hermann Frntratt ,

GPU-Accelerated Undecimated Wavelet Transform for Film and Video Denoising Hermann Frntratt ,

GPU-Accelerated Undecimated Wavelet Transform for Film and Video Denoising Hermann Frntratt , Hannes Fassold 2015-03-19 Overview 2 GPU-activities @ AVM research group Undecimated Wavelet Transform Algorithm GPU Implementation Application

300 views • 16 slides
The Haar Wavelet Transform: Compression and Adams and Halsey Reconstruction Patterson Damien

The Haar Wavelet Transform: Compression and Adams and Halsey Reconstruction Patterson Damien

The Haar Wavelet Transform: Compression and Recon- struction Damien The Haar Wavelet Transform: Compression and Adams and Halsey Reconstruction Patterson Damien Adams and Halsey Patterson December 13, 2006 The Haar Wavelet Transform:

338 views • 18 slides
Microcalcification Detection in Mammography using Wavelet Transform and Statistical Parameters

Microcalcification Detection in Mammography using Wavelet Transform and Statistical Parameters

Microcalcification Detection in Mammography using Wavelet Transform and Statistical Parameters Eliza Hashemi Supervisor: Alice Kozakevicius Examiner: Mohammad Assadzadeh Masters thesis presentation University of Gothenburg February 24,

828 views • 39 slides
Discrete Wavelet Transform Techniques for Denoising, Pattern Detection and Compression of

Discrete Wavelet Transform Techniques for Denoising, Pattern Detection and Compression of

Discrete Wavelet Transform Techniques for Denoising, Pattern Detection and Compression of Turbulent Rayleigh-Taylor Mix Data International Workshop on Bedros Afeyan, Polymath Research Inc. Praveen Ramaprabhu & Malcolm J. the Physics of

1.43k views • 130 slides
Non-linear Dynamics Characterization from Wavelet Packet Transform for Automatic Recognition of

Non-linear Dynamics Characterization from Wavelet Packet Transform for Automatic Recognition of

Non-linear Dynamics Characterization from Wavelet Packet Transform for Automatic Recognition of Emotional Speech J.C. Vsquez-Correa 1 , J.R Orozco-Arroyave 1,2 , J.D Arias-Londoo 1 , J.F Vargas-Bonilla 1 , Elmar Nth 2 1 Faculty of

232 views • 22 slides
Wavelet Scattering Transforms Haixia Liu Department of Mathematics The Hong Kong University of

Wavelet Scattering Transforms Haixia Liu Department of Mathematics The Hong Kong University of

Problem Wavelet Scattering Transform Digit Classification: MNIST by Joan Bruna et al. MATLAB code of Wavelet convolutional Networks Wavelet Scattering Transforms Haixia Liu Department of Mathematics The Hong Kong University of Science and

460 views • 24 slides
Continuous Wavelet Transform: ECG Recognition Based on Phase and Modulus Representations E DGAR G

Continuous Wavelet Transform: ECG Recognition Based on Phase and Modulus Representations E DGAR G

I NTRODUCTION T HE S QUARE M ODULUS AND P HASE E XAMPLES C ONCLUSION R EFERENCES Continuous Wavelet Transform: ECG Recognition Based on Phase and Modulus Representations E DGAR G ONZALEZ Based on the paper by: Lofti Senhadjii, Laurent Thoroval,

750 views • 64 slides
Continuous Wavelet Transform in Quantum Field Theory 1 Mikhail V. Altaisky 1 Natalia E.Kaputkina 2

Continuous Wavelet Transform in Quantum Field Theory 1 Mikhail V. Altaisky 1 Natalia E.Kaputkina 2

Continuous Wavelet Transform in Quantum Field Theory 1 Mikhail V. Altaisky 1 Natalia E.Kaputkina 2 1 Space Research Institute RAS, Profsoyuznaya 84/32, Moscow, 117997, Russia 2 National University of Science and Technology MISiS, Leninsky

869 views • 53 slides
Continuous Wavelet Transform and the Annual Cycle in Temperature and the Number of Deaths Milan

Continuous Wavelet Transform and the Annual Cycle in Temperature and the Number of Deaths Milan

Continuous Wavelet Transform and the Annual Cycle in Temperature and the Number of Deaths Milan Bata*, Josef Arlt* , , Markta Arltov*, , , Karel Helman* $ *Dept. of Statistics & Probability, Faculty of Informatics & Statistics,

207 views • 16 slides
Sparse signal representation and the tunable Q-factor wavelet transform Ivan Selesnick

Sparse signal representation and the tunable Q-factor wavelet transform Ivan Selesnick

Sparse signal representation and the tunable Q-factor wavelet transform Ivan Selesnick Polytechnic Institute of New York University Brooklyn, New York 1 Introduction Problem: Decomposition of a signal into the sum of two components: 1.

793 views • 62 slides
Regularization for Multi-Output Learning Lorenzo Rosasco 9.520 L. Rosasco Regularization for

Regularization for Multi-Output Learning Lorenzo Rosasco 9.520 L. Rosasco Regularization for

Regularization for Multi-Output Learning Lorenzo Rosasco 9.520 L. Rosasco Regularization for Multi-Output Learning About this class Goal In many practical problems, it is convenient to model the object of interest as a function with multiple

507 views • 47 slides
Regularization for Deep Learning Lecture slides for Chapter 7 of Deep Learning

Regularization for Deep Learning Lecture slides for Chapter 7 of Deep Learning

Regularization for Deep Learning Lecture slides for Chapter 7 of Deep Learning www.deeplearningbook.org Ian Goodfellow 2016-09-27 Definition Regularization is any modification we make to a learning algorithm that is intended to reduce

179 views • 13 slides
Some Essentials of Data Analysis with Wavelets Slides in the wavelet part of the course in data

Some Essentials of Data Analysis with Wavelets Slides in the wavelet part of the course in data

Some Essentials of Data Analysis with Wavelets Slides in the wavelet part of the course in data analysis at The Slid i h l f h i d l i Th Swedish National Graduate School of Space Technology Lecture 2: The continuous wavelet transform

808 views • 24 slides
Regularization Regularization is a general approach to add a complexity parameter to a

Regularization Regularization is a general approach to add a complexity parameter to a

Regularization Regularization is a general approach to add a complexity parameter to a learning algorithm. Requires that the model parameters be continuous. (i.e., Regression OK, IAML: Regularization and Ridge Regression Decision trees

204 views • 3 slides
CS7015 (Deep Learning) : Lecture 8 Regularization: Bias Variance Tradeoff, l2 regularization,

CS7015 (Deep Learning) : Lecture 8 Regularization: Bias Variance Tradeoff, l2 regularization,

CS7015 (Deep Learning) : Lecture 8 Regularization: Bias Variance Tradeoff, l2 regularization, Early stopping, Dataset augmentation, Parameter sharing and tying, Injecting noise at input, Ensemble methods, Dropout Mitesh M. Khapra Department of

1.06k views • 86 slides
Chem 342 Organic Chemistry II Spring 2004 c o o k . c h e m . n d s u . n o d a k . e d u / c h

Chem 342 Organic Chemistry II Spring 2004 c o o k . c h e m . n d s u . n o d a k . e d u / c h

Chem 342 Organic Chemistry II Spring 2004 c o o k . c h e m . n d s u . n o d a k . e d u / c h e m 3 4 2 Please pick up a syllabus near the entrance Syllabus Office Hours Mon, Wed - 9:00-10:00 am or give me a call Dunbar Hall 360A Phone -

452 views • 22 slides
Aging and Clinical Research What's In The Pipeline Objectives Learn about ongoing studies

Aging and Clinical Research What's In The Pipeline Objectives Learn about ongoing studies

Aging and Clinical Research What's In The Pipeline Objectives Learn about ongoing studies Think about which studies interest you most Know how to find trials near you! Snapshot Of Ongoing Research Exercise/Dance

207 views • 9 slides
Combined Image Reconstruction for Combined PET-MR Imaging Matthias J. Ehrhardt University of

Combined Image Reconstruction for Combined PET-MR Imaging Matthias J. Ehrhardt University of

Combined Image Reconstruction for Combined PET-MR Imaging Matthias J. Ehrhardt University of Cambridge, UK with: Arridge, Atkinson, Barnes, Duncan, Hutton, Liljeroth, Markiewicz Ourselin, Pizarro, Thielemans (UCL, London) Kolehmainen (Kuopio,

1.12k views • 87 slides
Visual Cognition Computer Vision 3D Vision Mobile multimedia 3D TV Model-based

Visual Cognition Computer Vision 3D Vision Mobile multimedia 3D TV Model-based

Visual Cognition Computer Vision 3D Vision Mobile multimedia 3D TV Model-based vision approaches An Imaginary experiment in Neuroimaging Biometrics Motion Analysis Color and texture Non-photorealistic

271 views • 3 slides
Promoting Mental Health and Well-being Best practices and lessons learned Workshop 2.1 Thursday,

Promoting Mental Health and Well-being Best practices and lessons learned Workshop 2.1 Thursday,

Promoting Mental Health and Well-being Best practices and lessons learned Workshop 2.1 Thursday, 19 October 2017; 16h30 18h30 Promoting Mental Health and Well-being Best practices and lessons learned Background: Shekhar Saxena, WHO Workshop

419 views • 21 slides
Depression, Anxiety, and Suicide Prevention Funded by SAMHSA in collaboration with AoA 1

Depression, Anxiety, and Suicide Prevention Funded by SAMHSA in collaboration with AoA 1

Depression, Anxiety, and Suicide Prevention Funded by SAMHSA in collaboration with AoA 1 Speakers Introductions & Welcome Jennifer Solomon Substance Abuse and Mental Health Administration Shannon Skowronski Administration

970 views • 64 slides
10Hz Repetitive Transcranial Magnetic Stimulation (rTMS) in Patients With Treatment Resistant

10Hz Repetitive Transcranial Magnetic Stimulation (rTMS) in Patients With Treatment Resistant

Cost Analysis of Intermittent Theta Burst Stimulation (iTBS) Versus 10Hz Repetitive Transcranial Magnetic Stimulation (rTMS) in Patients With Treatment Resistant Depression ANDREW B. MENDLOWITZ MBIOTECH, ALAA SHANBOUR MD, JONATHAN DOWNAR MD,

745 views • 43 slides
A Pathoplastic Presentation of Trichotillomania in Severe Depression- A Case Report Malalagama AS

A Pathoplastic Presentation of Trichotillomania in Severe Depression- A Case Report Malalagama AS

Journal of Psychiatry and Psychiatric Disorders doi: 10.26502/jppd.2572-519X0044 Case Report Volume 2, Issue 4 A Pathoplastic Presentation of Trichotillomania in Severe Depression- A Case Report Malalagama AS * Consultant Psychiatrist, Base

213 views • 4 slides

More recommend