Intelligent Assistant Diagnosis System of Osteosarcoma MRI Image Based on Transformer and Convolution in Developing...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Osteosarcoma diagnosis relies heavily on MRI, but image quality issues and large data volumes complicate manual interpretation, leading to potential misdiagnosis.
• Existing image segmentation models for osteosarcoma often overlook global features, limiting their diagnostic performance.

Purpose of the Study:

• To develop an intelligent assisted diagnosis system for osteosarcoma to alleviate the diagnostic burden on clinicians.
• To improve the accuracy and efficiency of osteosarcoma detection and segmentation from MRI scans.

Main Methods:

• A classification-image enhancement module (ResNet18 and DeepUPE) was developed to refine MRI quality and reduce redundant images.
• A novel Double U-shaped visual transformer with convolution (DUconViT) model was proposed for osteosarcoma segmentation, outperforming traditional convolutional methods.
• A pixel point quantification method was introduced for calculating osteosarcoma area.

Main Results:

• The DUconViT model achieved superior osteosarcoma segmentation performance, with Dice Similarity Coefficient (DSC) scores 2.6% and 1.8% higher than Unet and Unet++, respectively.
• The image enhancement module improved MRI clarity, facilitating better observation for clinicians.
• The pixel quantification method provided a quantitative basis for diagnosis.

Conclusions:

• The proposed intelligent assisted diagnosis system, featuring image enhancement and the DUconViT model, significantly aids in the accurate and efficient diagnosis of osteosarcoma.
• DUconViT demonstrates the potential of combining transformer and convolutional architectures for robust medical image segmentation.
• The system offers a valuable tool for reducing misdiagnosis rates and supporting clinical decision-making in osteosarcoma management.