Statistical Management of the False Discovery Rate in Medical Instance Segmentation Based on Conformal Risk Control

TL;DR

This work tackles the challenge of confidence calibration in medical instance segmentation by integrating conformal prediction with a calibration-aware loss to dynamically adjust segmentation thresholds at a user-defined risk level $α$. The proposed method is model-agnostic and compatible with common architectures like Mask R-CNN and BlendMask, requiring only a small exchangeable calibration set to guarantee that the expected loss on new data does not exceed $α$. The authors derive theoretical guarantees for FDR and FNR control, and validate robustness across varying calibration-to-test data ratios on a brain tumor MRI dataset, demonstrating controllable risk with practical computational efficiency. The approach enhances trustworthiness in high-stakes medical imaging, enabling reliable deployment of segmentation models in clinical workflows. Potential extensions include multi-instance segmentation and application to other medical imaging modalities.

Abstract

Instance segmentation plays a pivotal role in medical image analysis by enabling precise localization and delineation of lesions, tumors, and anatomical structures. Although deep learning models such as Mask R-CNN and BlendMask have achieved remarkable progress, their application in high-risk medical scenarios remains constrained by confidence calibration issues, which may lead to misdiagnosis. To address this challenge, we propose a robust quality control framework based on conformal prediction theory. This framework innovatively constructs a risk-aware dynamic threshold mechanism that adaptively adjusts segmentation decision boundaries according to clinical requirements.Specifically, we design a \textbf{calibration-aware loss function} that dynamically tunes the segmentation threshold based on a user-defined risk level $α$. Utilizing exchangeable calibration data, this method ensures that the expected FNR or FDR on test data remains below $α$ with high probability. The framework maintains compatibility with mainstream segmentation models (e.g., Mask R-CNN, BlendMask+ResNet-50-FPN) and datasets (PASCAL VOC format) without requiring architectural modifications. Empirical results demonstrate that we rigorously bound the FDR metric marginally over the test set via our developed calibration framework.

Figures (6)

• Figure 1: Visualization of Mask R-CNN FDR results. Red dashed line represents $\alpha$, blue line represents mean FDR loss, green line represents FNR loss.
• Figure 2: Visualization of BlendMask FDR results. Red dashed line represents $\alpha$, blue line represents mean FDR loss, green line represents FNR loss.
• Figure 3: Visualization of Mask R-CNN FNR results. Red dashed line represents $\alpha$, blue line represents mean FNR loss, green line represents FDR loss.
• Figure 4: Visualization of BlendMask FNR results. Red dashed line represents $\alpha$, blue line represents mean FNR loss, green line represents FDR loss.
• Figure 5: Ablation study results for Mask R-CNN. Blue line represents FDR loss, orange line represents FNR loss, red dashed line represents fixed $\alpha=0.25$.