YM-WML: A new Yolo-based segmentation Model with Weighted Multi-class Loss for medical imaging

TL;DR

This paper introduces YM-WML, a novel YOLO-based model with a weighted multi-class loss for improved cardiac image segmentation, addressing class imbalance and complex structures, and achieving state-of-the-art results.

Contribution

The paper presents YM-WML, a new segmentation model combining YOLOv11 architecture and a weighted loss function, advancing cardiac image segmentation performance.

Findings

Achieved 91.02 Dice coefficient on ACDC dataset.

Outperformed existing state-of-the-art methods.

Demonstrated stable training and strong generalization.

Abstract

Medical image segmentation poses significant challenges due to class imbalance and the complex structure of medical images. To address these challenges, this study proposes YM-WML, a novel model for cardiac image segmentation. The model integrates a robust backbone for effective feature extraction, a YOLOv11 neck for multi-scale feature aggregation, and an attention-based segmentation head for precise and accurate segmentation. To address class imbalance, we introduce the Weighted Multi-class Exponential (WME) loss function. On the ACDC dataset, YM-WML achieves a Dice Similarity Coefficient of 91.02, outperforming state-of-the-art methods. The model demonstrates stable training, accurate segmentation, and strong generalization, setting a new benchmark in cardiac segmentation tasks.