Transfer Learning for Cell Nuclei Classification in Histopathology - - PowerPoint PPT Presentation

▶

Jul 28, 2023 357 likes •467 views

Transfer Learning for Cell Nuclei Classification in Histopathology Images Neslihan Bayramoglu, Janne Heikkil Center for Machine Vision and Signal Analysis, University of Oulu, Finland. Biomedical Image Analysis Histopathological image

SLIDE 1

Transfer Learning for Cell Nuclei Classification in Histopathology Images

Neslihan Bayramoglu, Janne Heikkilä Center for Machine Vision and Signal Analysis, University of Oulu, Finland.

SLIDE 2

Neslihan Bayramoglu Transfer Learning for Cell Nuclei Classification in Histopathology Images

Histopathological image assessment
microscopic examination of tissue
High demand to obtain fast and precise quantification

automatically.

Adenoid cystic carcinoma Ductal carcinoma of the breast Ovarian cancer Stained with Hematoxylin & Eosin (H&E)

Biomedical Image Analysis

SLIDE 3

Neslihan Bayramoglu Transfer Learning for Cell Nuclei Classification in Histopathology Images

Automated techniques beneficial to
find clinical assessment clues,
produce correct diagnoses,
reduce observer variability,
increase objectivity.
Deep learning could be the key method to obtain clinical

acceptance

Biomedical Image Analysis

SLIDE 4

Neslihan Bayramoglu Transfer Learning for Cell Nuclei Classification in Histopathology Images

Deep Learning for Biomedical Image Analysis

Bottleneck:

Limited amount of training data.

Question:

Could it be possible to use transfer learning and fine-tuning in biomedical image analysis to reduce the effort of manual data labeling and still obtain a full deep representation for the target task?

SLIDE 5

Neslihan Bayramoglu Transfer Learning for Cell Nuclei Classification in Histopathology Images

Significant differences in image statistics between biomedical

images and natural images

We evaluate whether the features learned from deep CNNs

trained on generic recognition tasks could generalize to biomedical tasks.

Deep Learning for Biomedical Image Analysis

SLIDE 6

Neslihan Bayramoglu Transfer Learning for Cell Nuclei Classification in Histopathology Images

Significant differences in image statistics between biomedical

images and natural images

We evaluate whether the features learned from deep CNNs

trained on generic recognition tasks could generalize to biomedical tasks.

Deep Learning for Biomedical Image Analysis

SLIDE 7

Neslihan Bayramoglu Transfer Learning for Cell Nuclei Classification in Histopathology Images

Data Set

H&E stained histopathology images of colorectal adenocarcinoma.
20,405 manually labelled cell nuclei (training: 17,004, testing : 3401).
Categories: Epithelial (7,772), inflammatory (6,971), fibroblast (5,712),

and miscellaneous (excluded).

Publicy available.

Cell Nuclei Classification in Histopathology Images

SLIDE 8

Neslihan Bayramoglu Transfer Learning for Cell Nuclei Classification in Histopathology Images

AlexNet 5 Layers GenderNet 3 Layers VGG-16 13 Layers GoogLeNet 22 Layers

Experiments

Full training: network parameters are initialized randomly.
Raw images, no data augmentation.

SLIDE 9

Neslihan Bayramoglu Transfer Learning for Cell Nuclei Classification in Histopathology Images

0.7 0.8 0.9

VGG_16 Number of Iterations

500 1,000 1,500 2,000 2,500 3,000 0.5 0.6 0.7 0.8 0.9

GoogLeNet

Number of Iterations 200 400 600 800 1,000 1,200 1,400 1,600 0.7 0.8 0.9

GenderNet

Transfer Learning Full Training

Classification Accuracy 0.6 0.7 0.8 0.9

AlexNet

Number of Iterations 200 400 600 800 1,000 1,200 1,400 1,600 Classification Accuracy

Results

Transfer learning and fine-tuning provides much better results than learning from scratch. Initializing the network parameters with transferred features can improve the classification performance for any model. Deeper architectures trained on bigger datasets converge faster. Feature transferability is affected by the depth of the network, source task, and the diversity of the source data.