Differentiating spinal pathologies by deep learning approach.

Area of Science:

• Artificial Intelligence in Medicine
• Spinal Pathology Diagnostics
• Medical Imaging Analysis

Background:

• Spinal pathologies encompass diverse conditions including infectious, neoplastic, and inflammatory diseases.
• Accurate preoperative diagnosis is crucial for effective treatment, often requiring invasive and costly biopsies when clinical and radiological data are inconclusive.

Purpose of the Study:

• To develop a deep learning (DL) algorithm capable of differentiating between major spinal pathologies.
• The algorithm leverages preoperative Magnetic Resonance Imaging (MRI) and post-operative pathological findings.

Main Methods:

• Retrospective analysis of clinical, radiological, and pathological data from 231 patients (2008-2022).
• A deep learning model was constructed using the Fast.ai framework and PyTorch, processing preoperative MRI, clinical data, and pathological results.
• Patients were stratified by diagnosis, with a minimum of 25 cases per pathology.

Main Results:

• The DL model achieved a mean accuracy of 0.78±0.06 on the validation dataset and 0.93±0.03 on the test dataset.
• The model successfully differentiated between carcinoma, infection, meningioma, and schwannoma.

Conclusions:

• Deep learning algorithms utilizing solely preoperative MRI show promise as a primary diagnostic tool for spinal pathologies.
• Further validation and larger studies are recommended to refine the algorithm for widespread clinical application.
• This technology offers a potential for rapid, targeted, and cost-effective patient work-up and treatment planning.