Feature Disentanglement to Aid Imaging Biomarker Characterization for Genetic Mutations

Padmaja Jonnalagedda, Brent Weinberg, Jason Allen, Bir Bhanu

Keywords: Generative Adversarial Networks, Feature Disentanglement, Brain Tumor, Magnetic Resonance Imaging, Limited Dataset

TL;DR: A framework to detect and synthesize genetic alterations in medical imaging using limited dataset by splitting and resampling of features

Track: full conference paper

Paper Type: both

Abstract: Various mutations have been shown to correlate with prognosis of High-Grade Glioma (Glioblastoma). Overall prognostic assessment requires analysis of multiple modalities: imaging, molecular and clinical. To optimize this assessment pipeline, this paper develops the first deep learning-based system that uses MRI data to predict 19/20 co-gain, a mutation that indicates median survival. It addresses two key challenges when dealing with deep learning algorithms and medical data: lack of data and high data imbalance. To tackle these challenges, we propose a unified approach that consists of a Feature Disentanglement based Generative Adversarial Network (FeaD-GAN) for generating synthetic images. FeaD-GAN projects disentangled features into a high dimensional space and re-samples them from a pseudo-large data distribution to generate synthetic images from very limited data. A thorough analysis is performed to (a) characterize aspects of visual manifestation of 19/20 co-gain to demonstrate the effectiveness of FeaD-GAN and (b) demonstrate that not only do the imaging biomarkers of 19/20 co-gain exist, but also that they are reproducible.

Presentation Upload Agreement: I agree that my presentation material (videos and slides) will be made public.