CALF: A Conditionally Adaptive Loss Function to Mitigate Class-Imbalanced Segmentation

2025-04-06

AI Summary

In medical image segmentation, poor annotation quality, sparse lesions, and minute regions of interest (ROIs) induce severe class imbalance, causing conventional loss functions (e.g., binary cross-entropy, BCE) to amplify annotation bias and degrade boundary localization accuracy. To address this, we propose the Conditional Adaptive Loss Function (CALF), the first to quantify imbalance severity via skewness and kurtosis statistics and dynamically select or weight loss components accordingly—thereby jointly modeling data heterogeneity and correcting imbalance. CALF integrates statistical modeling, data-driven preprocessing, and dynamic loss scheduling. Evaluated across multi-center datasets—including UPENN-GBM, UCSF, LGG, and BraTS—CALF consistently improves segmentation accuracy for small targets and enhances boundary precision, outperforming BCE and other mainstream baselines.

Technology Category

Machine Learning: Calibration & Uncertainty QuantificationComputer Vision: SegmentationReasoning under Uncertainty: Graphical Models

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingGraph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphsUser Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Imbalanced datasets pose a considerable challenge in training deep learning (DL) models for medical diagnostics, particularly for segmentation tasks. Imbalance may be associated with annotation quality limited annotated datasets, rare cases, or small-scale regions of interest (ROIs). These conditions adversely affect model training and performance, leading to segmentation boundaries which deviate from the true ROIs. Traditional loss functions, such as Binary Cross Entropy, replicate annotation biases and limit model generalization. We propose a novel, statistically driven, conditionally adaptive loss function (CALF) tailored to accommodate the conditions of imbalanced datasets in DL training. It employs a data-driven methodology by estimating imbalance severity using statistical methods of skewness and kurtosis, then applies an appropriate transformation to balance the training dataset while preserving data heterogeneity. This transformative approach integrates a multifaceted process, encompassing preprocessing, dataset filtering, and dynamic loss selection to achieve optimal outcomes. We benchmark our method against conventional loss functions using qualitative and quantitative evaluations. Experiments using large-scale open-source datasets (i.e., UPENN-GBM, UCSF, LGG, and BraTS) validate our approach, demonstrating substantial segmentation improvements. Code availability: https://anonymous.4open.science/r/MICCAI-Submission-43F9/.

Problem

• Mitigates class-imbalanced segmentation in medical diagnostics
• Addresses annotation biases in deep learning models
• Improves segmentation accuracy for rare or small ROIs

Innovation

• Conditionally adaptive loss function CALF
• Estimates imbalance via skewness and kurtosis
• Dynamic loss selection improves segmentation