Enhanced Detection, Using Deep Learning Technology, of Medial Meniscal Posterior Horn Ramp Lesions in Patients with...

Area of Science:

• Orthopedic surgery
• Radiology
• Artificial intelligence in medicine

Background:

• Meniscal ramp lesions are crucial in knee stability, often co-occurring with anterior cruciate ligament (ACL) injuries.
• Magnetic resonance imaging (MRI) is the standard for diagnosis, but its accuracy for ramp lesions is limited.
• Deep learning (DL) offers potential to enhance diagnostic capabilities for these subtle injuries.

Purpose of the Study:

• To evaluate the efficacy of a DL model in improving the detection of meniscal ramp lesions using MRI data.
• To assess the added value of integrating clinical risk factors into the DL model for enhanced diagnostic performance.

Main Methods:

• A DL model was developed using MRI scans from 236 patients with ACL injuries.
• Ramp lesion risk factors were identified using logistic regression, XGBoost, and random forest models.
• A Swin Transformer Large architecture integrated MRI data and risk factors for a final prediction model.

Main Results:

• The DL model showed higher accuracy (73.3%) than clinicians (68.1%) using MRI alone.
• Incorporating risk factors (age, bone marrow edema, lateral meniscal tears) boosted model accuracy to 80.7%, with improved sensitivity and specificity.

Conclusions:

• Integrating DL with MRI and clinical risk factors significantly enhances diagnostic accuracy for meniscal ramp lesions.
• This AI-driven approach surpasses traditional methods and clinician assessments, offering a promising tool for improved diagnosis and patient outcomes.