Multiple supervised residual network for osteosarcoma segmentation in CT images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate segmentation of osteosarcoma in CT images is crucial for effective treatment planning.
• Existing segmentation methods may lack the precision required for optimal clinical decision-making.

Purpose of the Study:

• To develop and evaluate a novel deep learning model, the Multiple Supervised Residual Network (MSRN), for precise osteosarcoma segmentation in CT images.
• To compare the performance of MSRN against established segmentation techniques like Fully Convolutional Network (FCN) and U-Net.

Main Methods:

• Proposed a Multiple Supervised Residual Network (MSRN) incorporating three supervised side output modules within a residual network architecture.
• Shallow modules extracted shape and texture features; deep modules extracted semantic features.
• Integrated multi-scale feature learning guided by back-propagated loss information from side outputs.

Main Results:

• MSRN achieved a Dice Similarity Coefficient (DSC) of 89.22%, sensitivity of 88.74%, and F1-measure of 0.9305 on a dataset of 405 CT images.
• These performance metrics surpassed those of FCN and U-Net.
• The network was trained on 1900 CT images from 15 patients and tested on 405 images from 8 patients.

Conclusions:

• The MSRN model demonstrates superior accuracy in osteosarcoma segmentation compared to FCN and U-Net.
• This advanced segmentation technique holds significant potential for improving diagnostic accuracy and treatment planning in osteosarcoma management.