Defense (PhD, Masters, Pre)

Masters Defense

Improvements in cross-modality cardiac segmentation for unsupervised domain adaptation frameworks

 

Download as iCal file

Wednesday, March 18, 2020, 03:00pm - 04:30pm

 

Speaker: Rushin Gindra

Location : CoRE 301

Committee

Prof. Dimitris N. Metaxas (Advisor)

Prof. Konstantinos Michmizos

Prof. Karl Stratos

Event Type: Masters Defense

Abstract: In medical image computing, the problem of heterogeneous domain shift is quite common and severe, causing many deep convolutional networks to under-perform on various imaging modalities. Retraining the network is difficult since annotating the new domain data is prohibitively expensive, specifically in medical areas that require expertise. While recent works show approaches to tackle this problem using unsupervised domain adaptation, segmentation modules in such methods can be improved vastly. Our implementation provides two segmentation improvements on the current state-of-the-art framework, Synergistic Image and Feature Adaptation(SIFA). We revisit cascade multi-rate atrous convolutions and also borrow the idea of atrous spatial pyramid pooling while using convolutional features as well as image features for multi-scale object segmentation. We have validated the effectiveness of the improvements on the framework using the challenging application of cross-modality segmentation of cardiac structures. To demonstrate the robustness of the module, extensive experiments have been performed on both Long-Axis(MMWHS) & Short-Axis(MSCMR) cross-modal cardiac segmentation tasks.