Development and validation of a multi-modal MRI-based deep learning framework for differentiation of intraspinal...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Intraspinal tumors (ISTs) present diagnostic challenges due to complex anatomy and reliance on subjective MRI interpretation.
• Artificial intelligence (AI) offers potential for non-invasive IST diagnosis.
• Developing AI models for IST differentiation is crucial for improving patient outcomes.

Purpose of the Study:

• To develop and validate an MRI-based deep learning model for differentiating various types of ISTs.
• To assess the diagnostic performance of the proposed model.
• To evaluate the impact of AI assistance on radiologist diagnostic performance.

Main Methods:

• A retrospective study included 1004 patients with ISTs (Schwannoma, Meningioma, Astrocytoma, Ependymoma, Metastasis) across three hospitals.
• A deep learning model, ISMF-Net, was developed integrating multi-sequence MRI and clinical data.
• The model was validated on internal and external test sets, and an observer study assessed diagnostic improvements.

Main Results:

• ISMF-Net achieved high diagnostic accuracy (micro-ACC ~0.82-0.86) across validation, internal, and external test sets.
• The model demonstrated superior performance compared to existing methods, with notable success in differentiating Schwannoma.
• AI assistance significantly improved radiologist accuracy, sensitivity, and specificity, particularly for junior radiologists.

Conclusions:

• The proposed multi-modal deep learning approach (ISMF-Net) effectively integrates MRI and clinical data for IST diagnosis.
• ISMF-Net serves as a valuable tool for enhancing diagnostic accuracy in clinical practice.
• AI-assisted diagnosis holds significant promise for improving the management of intraspinal tumors.