Do We Really Need Dice? The Hidden Region-Size Biases of Segmentation Losses

Abstract

Most segmentation losses are arguably variants of the Cross-Entropy (CE) or Dice losses. On the surface, these two categories of losses seem unrelated, and there is no clear consensus as to which category is a better choice, with varying performances for each across different benchmarks and applications. Furthermore, it is widely argued within the medical-imaging community that Dice and CE are complementary, which has motivated the use of compound CE-Dice losses. In this work, we provide a theoretical analysis, which shows that CE and Dice share a much deeper connection than previously thought. First, we show that, from a constrained-optimization perspective, they both decompose into two components, i.e., a similar ground-truth matching term, which pushes the predicted foreground regions towards the ground-truth, and a region-size penalty term imposing different biases on the size of the predicted regions. Then, we provide bound relationships and an information-theoretic analysis, which uncover hidden region-size biases: Dice has an intrinsic bias towards specific extremely imbalanced solutions, whereas CE implicitly encourages the ground-truth region proportions. Our theoretical results explain the wide experimental evidence in the medical-imaging literature, whereby Dice losses bring improvements for imbalanced segmentation. Based on our theoretical analysis, we propose a principled and simple solution, which enables to control explicitly the region-size bias. The proposed method integrates CE with explicit terms based on L1 or the KL divergence, which encourage segmenting region proportions to match target class proportions, thereby mitigating class imbalance but without losing generality. Comprehensive experiments and ablation studies over different losses and applications validate our theoretical analysis, as well as the effectiveness of explicit and simple region-size terms.

To be published at Medical Image Analysis(Volume 91, January 2024, 103015). Code available at https://github.com/by-liu/SegLossBias

References in corpus (8)

• The Medical Segmentation Decathlon
• BEiT: BERT Pre-Training of Image Transformers
• Optimization for Medical Image Segmentation: Theory and Practice when evaluating with Dice Score or Jaccard Index
• Hierarchical Multi-Scale Attention for Semantic Segmentation
• Monte-Carlo Sampling applied to Multiple Instance Learning for Histological Image Classification
• AMOS: A Large-Scale Abdominal Multi-Organ Benchmark for Versatile Medical Image Segmentation
• PolyLoss: A Polynomial Expansion Perspective of Classification Loss Functions
• blob loss: instance imbalance aware loss functions for semantic segmentation

Cited by in corpus (1)

• Towards Investigating Residual Hearing Loss: Quantification of Fibrosis in a Novel Cochlear OCT Dataset