Multi-Level Seg-Unet Model with Global and Patch-Based X-ray Images for Knee Bone Tumor Detection

Area of Science:

• Medical imaging analysis
• Artificial intelligence in oncology
• Orthopedic oncology

Background:

• Deep learning applications for tumor classification and segmentation are prevalent in brain, lung, and liver cancers.
• There is a notable gap in deep learning research for classifying and segmenting knee bone tumors.

Purpose of the Study:

• To develop a deep learning model to aid physicians in radiographic interpretation of knee bone tumors.
• To classify knee bone regions as normal, benign-tumor, or malignant-tumor.
• To segment tumorous regions within knee bone radiographs.

Main Methods:

• Proposed the Seg-Unet model incorporating both global and patch-based approaches.
• Developed classification, tumor segmentation, and high-risk region segmentation branches.
• Utilized a dataset of knee bone tumors supported by Chonnam National University Hospital (CNUH) physicians.

Main Results:

• Achieved 99.05% accuracy for tumor classification.
• Obtained an average Mean Intersection over Union (IoU) of 84.84% for tumor segmentation.
• Demonstrated superior performance compared to existing methods, particularly in malignant-tumor detection via the patch-based approach.

Conclusions:

• The Seg-Unet model effectively classifies and segments knee bone tumors.
• The combined global and patch-based approach addresses challenges like small lesion size and varied appearance.
• This AI-driven tool shows significant potential to assist physicians in diagnosing knee bone tumors.