4D Deep Learning for Multiple Sclerosis Lesion Activity Segmentation - PowerPoint PPT Presentation

Nils Gessert 1 , Marcel Bengs 1 , Julia Krüger 2 , Roland Opfer 2 , Ann-Christin Ostwaldt 2 , Praveena Manogaran 3 , Sven Schippling 3 , Alexander Schlaefer 1 Institute of Medical Technology and Intelligent Systems 4D Deep Learning for Multiple Sclerosis Lesion Activity Segmentation 1 Institute of Medical Technology and Intelligent Systems, Hamburg University of Technology 2 jung diagnostics GmbH 3 Department of Neurology, University Hospital Zurich and University of Zurich

July 2020 Slide 2 Lesion Activity Segmentation Enlarged Lesion Enlargement Old Lesion Follow-up MRI Scan Baseline MRI Scan Lesion Activity

July 2020 Slide 3 Two-Path 3D Encoder-Decoder Baseline History Follow-Up & Prediction Follow-Up Krüger, Julia, et al. "Fully automated longitudinal segmentation of new or enlarging Multiple Scleroses (MS) lesions using 3D convolution neural networks."

July 2020 Slide 4 Encoder-convGRU-Decoder Dataset: • 44 MS cases, three time points each • FLAIR image volumes, varying size

July 2020 Slide 5 Results and Discussion 65 84,5 40 84 64 35 83,5 63 30 83 62 25 82,5 61 82 20 60 81,5 15 59 81 10 58 80,5 5 57 80 56 79,5 0 Dice LTPR LFPR Enc-Dec T=2 Enc-Dec T=3 Enc-Dec T=2 Enc-Dec T=3 Enc-Dec T=2 Enc-Dec T=3 Enc-cGRU-Dec T=2 Enc-cGRU-Dec T=3 Enc-cGRU-Dec T=2 Enc-cGRU-Dec T=3 Enc-cGRU-Dec T=2 Enc-cGRU-Dec T=3