NMTNet: A Multi-task Deep Learning Network for Joint Segmentation and Classification of Breast Tumors

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate segmentation and classification of breast tumors are crucial for computer-aided diagnosis.
• Tumor variability presents challenges for existing methods.
• Jointly addressing segmentation and classification can enhance performance.

Purpose of the Study:

• To propose a novel multi-task deep learning network (NMTNet) for the joint segmentation and classification of breast tumors.
• To improve the accuracy and robustness of breast tumor analysis in medical imaging.

Main Methods:

• Developed a novel multi-task deep learning network (NMTNet) using a CNN and U-shaped architecture.
• Incorporated a shared encoder (ResNet18), multi-scale fusion channel refinement (MFCR) module, and lesion region enhancement (LRE) module.
• Integrated segmentation and classification branches, with the classifier leveraging segmentation information.

Main Results:

• NMTNet achieved high performance on ultrasound and MRI datasets.
• Segmentation Dice scores reached 90.30% and 91.50%, Jaccard indices 84.70% and 88.10%.
• Classification accuracy reached 87.50% and 99.64% for the respective datasets.

Conclusions:

• NMTNet demonstrates superior performance compared to state-of-the-art methods for breast tumor segmentation and classification.
• The proposed network effectively handles tumor variability and complexity.
• NMTNet shows significant potential for advancing computer-aided breast cancer diagnosis.