Bayesian approaches for Quantifying Clinicians' Variability in Medical Image Quantification

TL;DR

This paper investigates whether Bayesian deep learning can effectively model and quantify the variability among clinicians in medical image segmentation, providing a way to incorporate uncertainty into clinical measurements.

Contribution

It demonstrates that Bayesian neural networks can approximate clinicians' inter- and intra-rater variability in segmentation and measurements across MRI and ultrasound images.

Findings

Bayesian models capture clinician variability effectively.

Uncertainty quantification improves clinical measurement reliability.

Applicable to multiple imaging modalities.

Abstract

Medical imaging, including MRI, CT, and Ultrasound, plays a vital role in clinical decisions. Accurate segmentation is essential to measure the structure of interest from the image. However, manual segmentation is highly operator-dependent, which leads to high inter and intra-variability of quantitative measurements. In this paper, we explore the feasibility that Bayesian predictive distribution parameterized by deep neural networks can capture the clinicians' inter-intra variability. By exploring and analyzing recently emerged approximate inference schemes, we evaluate whether approximate Bayesian deep learning with the posterior over segmentations can learn inter-intra rater variability both in segmentation and clinical measurements. The experiments are performed with two different imaging modalities: MRI and ultrasound. We empirically demonstrated that Bayesian predictive distribution parameterized by deep neural networks could approximate the clinicians' inter-intra variability. We show a new perspective in analyzing medical images quantitatively by providing clinical measurement uncertainty.