Eleni Chiou - Adam Szmul - CMIC/WEISS joint seminar series

Eleni Chiou

Title: Harnessing Uncertainty in Domain Adaptation for Prostate Lesion Segmentation on VERDICT-MRI

Abstract

The need for training data can impede the adoption of novel imaging modalities for learning-based medical image analysis. Domain adaptation methods partially mitigate this problem by translating training data from a related source domain to a novel target domain, but typically assume a one-to-one translation is possible. Our work addresses the challenge of adapting to a more informative target domain where multiple target samples can emerge from a single source sample. In particular we consider translating from mp-MRI to VERDICT, a richer MRI modality that captures microstructural information. We explicitly account for the inherent uncertainty of this mapping and exploit it to generate multiple outputs conditioned on a single input. Our results show that this allows us to extract systematically better image representations for the target domain, when used in tandem with both simple, CycleGAN-based baselines, as well as more powerful approaches that integrate discriminative segmentation losses and/or residual adapters. When compared to its deterministic counterparts, our approach yields substantial improvements

across a broad range of dataset sizes, and evaluation measures.

Adam Szmul

Title: Correlation analysis between radiation induced lung changes and dose volume histograms in lung cancer follow-up

Abstract:

Lung related diseases are one of the leading causes of death worldwide, with lung cancer being the most common cause of cancer death. One effective treatment exposes the cancer cells to high energy radiation, which is known as radiotherapy. Radiation Induced Lung Damage (RILD) is a side effect of this treatment and one of the main factors reducing quality of life of cancer survivors. Our group has developed and presented a set of objective imaging Biomarkers suitable for quantification of RILD. Here we apply them on lung cancer 12m follow-up cohort. The Biomarkers' relationship with dosimetry and currently used clinical measures, such as MRC scores and pulmonary function tests, has not yet been fully established. We present such an analysis where we investigate the correlations between the RILD Biomarkers, PFTs, MRCs, and dosimetry represented in a form of cumulative and differential Dose Volume Histograms. Our results show similar level of correlation between the RILD biomarkars and dosimetry as for PFTs/MRCs and dosimetry. Finally, we present the ongoing work, where we extend the RILD analysis to parenchymal changes, which has the potential to provide complementary information about the RILD to the currently used RILD Biomarkers.