Artificial Intelligence-Aided Diagnosis Solution by Enhancing the Edge Features of Medical Images

Area of Science:

• Medical imaging analysis
• Artificial intelligence in oncology
• Biomedical engineering

Background:

• Manual screening of osteosarcoma MRI images is time-consuming and challenging due to noise and blurred edges.
• Existing high-precision segmentation methods demand significant computational resources, limiting accessibility in resource-constrained settings.
• Accurate segmentation is crucial for effective diagnosis and treatment planning of aggressive bone malignant tumors.

Purpose of the Study:

• To develop an efficient and accessible artificial intelligence-aided diagnosis scheme for osteosarcoma.
• To enhance the segmentation of osteosarcoma lesions in MRI images, particularly addressing issues with blurred edges and noise.
• To improve diagnostic efficiency and provide better support for clinicians in diagnosing osteosarcoma.

Main Methods:

• A threshold screening filter (TSF) was employed for pre-screening MRI images and filtering redundant data.
• A fast Non-Local Means (NLM) algorithm was utilized for image denoising.
• A novel segmentation method, TBNet, integrating a Transformer with a U-Net architecture and a combined loss function, was designed for edge-enhanced segmentation.

Main Results:

• The proposed TBNet method demonstrated a strong segmentation performance on over 4000 osteosarcoma MRI images.
• The Dice Similarity Coefficient (DSC) achieved was 0.949, indicating high accuracy.
• Other evaluation metrics, including Intersection of Union (IOU) and recall, showed superior results compared to existing methods.

Conclusions:

• The developed AI-aided diagnosis scheme effectively addresses the challenges of noise and blurred edges in osteosarcoma MRI segmentation.
• TBNet optimizes diagnostic efficiency and provides a valuable tool for clinicians, enhancing osteosarcoma diagnosis.
• The method offers a practical and high-performing solution, suitable for wider adoption, including in developing countries.