On the relationship between calibrated predictors and unbiased volume estimation

TL;DR

This paper explores how calibrated predictors in medical image segmentation can be used to accurately estimate volumes, demonstrating that calibration per image leads to unbiased volume estimates, with implications for medical applications.

Contribution

It establishes a mathematical and empirical link between calibration and unbiased volume estimation, showing calibration per image suffices for accurate volume measurement.

Findings

Calibrated per-image predictors enable correct volume estimation.

Convex combinations of calibrated classifiers preserve volume estimates.

Calibration is sufficient but not necessary for unbiased volume estimation.

Abstract

Machine learning driven medical image segmentation has become standard in medical image analysis. However, deep learning models are prone to overconfident predictions. This has led to a renewed focus on calibrated predictions in the medical imaging and broader machine learning communities. Calibrated predictions are estimates of the probability of a label that correspond to the true expected value of the label conditioned on the confidence. Such calibrated predictions have utility in a range of medical imaging applications, including surgical planning under uncertainty and active learning systems. At the same time it is often an accurate volume measurement that is of real importance for many medical applications. This work investigates the relationship between model calibration and volume estimation. We demonstrate both mathematically and empirically that if the predictor is calibrated per image, we can obtain the correct volume by taking an expectation of the probability scores per pixel/voxel of the image. Furthermore, we show that convex combinations of calibrated classifiers preserve volume estimation, but do not preserve calibration. Therefore, we conclude that having a calibrated predictor is a sufficient, but not necessary condition for obtaining an unbiased estimate of the volume. We validate our theoretical findings empirically on a collection of 18 different (calibrated) training strategies on the tasks of glioma volume estimation on BraTS 2018, and ischemic stroke lesion volume estimation on ISLES 2018 datasets.