Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Beyond Traditional Approaches: Multi-Task Network for Breast Ultrasound Diagnosis

Dat T. Chung, Minh-Anh Dang, Mai-Anh Vu, Minh T. Nguyen, Thanh-Huy Nguyen, Vinh Q. Dinh

TL;DR

This work tackles the need for simultaneous segmentation and classification in breast ultrasound imaging. It proposes an end-to-end multi-task network with a shared backbone and two task-specific heads, evaluated across backbones such as ResNet-50, ResNeXt-50, WideResNet-50, and EfficientNet-B4, with decoders including Unet, Unet++, and DeepLabV3+. The model optimizes a joint objective by combining segmentation and classification losses, with a weighting parameter and loss components focusing on Dice Loss for segmentation and Focal Loss for classification; best performance occurs at lambda = 0.7, achieving up to 86.4% segmentation metric in the DeepLabV3+ variant. The results suggest that multi-task learning can enhance boundary delineation and tumor attribute prediction while potentially reducing inference time and resource needs, supporting practical clinical deployment in breast cancer screening.

Abstract

Breast Ultrasound plays a vital role in cancer diagnosis as a non-invasive approach with cost-effective. In recent years, with the development of deep learning, many CNN-based approaches have been widely researched in both tumor localization and cancer classification tasks. Even though previous single models achieved great performance in both tasks, these methods have some limitations in inference time, GPU requirement, and separate fine-tuning for each model. In this study, we aim to redesign and build end-to-end multi-task architecture to conduct both segmentation and classification. With our proposed approach, we achieved outstanding performance and time efficiency, with 79.8% and 86.4% in DeepLabV3+ architecture in the segmentation task.

Beyond Traditional Approaches: Multi-Task Network for Breast Ultrasound Diagnosis

TL;DR

This work tackles the need for simultaneous segmentation and classification in breast ultrasound imaging. It proposes an end-to-end multi-task network with a shared backbone and two task-specific heads, evaluated across backbones such as ResNet-50, ResNeXt-50, WideResNet-50, and EfficientNet-B4, with decoders including Unet, Unet++, and DeepLabV3+. The model optimizes a joint objective by combining segmentation and classification losses, with a weighting parameter and loss components focusing on Dice Loss for segmentation and Focal Loss for classification; best performance occurs at lambda = 0.7, achieving up to 86.4% segmentation metric in the DeepLabV3+ variant. The results suggest that multi-task learning can enhance boundary delineation and tumor attribute prediction while potentially reducing inference time and resource needs, supporting practical clinical deployment in breast cancer screening.

Abstract

Breast Ultrasound plays a vital role in cancer diagnosis as a non-invasive approach with cost-effective. In recent years, with the development of deep learning, many CNN-based approaches have been widely researched in both tumor localization and cancer classification tasks. Even though previous single models achieved great performance in both tasks, these methods have some limitations in inference time, GPU requirement, and separate fine-tuning for each model. In this study, we aim to redesign and build end-to-end multi-task architecture to conduct both segmentation and classification. With our proposed approach, we achieved outstanding performance and time efficiency, with 79.8% and 86.4% in DeepLabV3+ architecture in the segmentation task.
Paper Structure (16 sections, 3 equations, 3 figures, 1 table)

This paper contains 16 sections, 3 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related works
  3. Medical Image Segmentation
  4. Multi-task learning
  5. Methodology
  6. Multi-task framework
  7. Classification head
  8. Segmentation head
  9. Loss composition
  10. Classification
  11. Segmentation
  12. Experiments Results
  13. Experimental Setting.
  14. Overall Results
  15. Visualization
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Visualization of our proposed architecture with multi-task learning approach. The shared encoder (green area) takes weighted sum loss from the segmentation branch (blue area) and classification branch (yellow area).
  • Figure 2: F1-score and Overall score of segmentation and classification task
  • Figure 3: Visualization of prediction results of the naive approach and our multi-task framework. (a) Breast ultrasound image; (b) Ground truth masks; (c) Naive segmentation approach output; (d) Our multi-task framework output.