Multimodal Deep Learning-based Radiomics Approach for Predicting Surgical Outcomes in Patients with Cervical...

Area of Science:

• Spinal Surgery
• Machine Learning in Medicine
• Predictive Analytics

Background:

• Surgical outcomes are crucial for patient prognosis and expectation management in cervical ossification of the posterior longitudinal ligament (OPLL).
• Deep learning and machine learning (ML) offer powerful tools for identifying patterns in large datasets to predict outcomes.

Purpose of the Study:

• To develop a predictive model for surgical outcomes in cervical ossification of the posterior longitudinal ligament (OPLL) patients.
• Utilize deep learning and machine learning (ML) techniques for outcome prediction.

Main Methods:

• A retrospective analysis of 288 patients with cervical ossification of the posterior longitudinal ligament (OPLL).
• Developed a predictive model using LightGBM and deep learning with RadImagenet.
• Incorporated patient background, clinical symptoms, and preoperative imaging (x-ray, CT, MRI) to predict minimal clinically important difference (MCID) achievement at 1 year.

Main Results:

• 60.1% of patients achieved the minimal clinically important difference (MCID) at 1 year post-surgery.
• The predictive model demonstrated an area under the curve of 0.81 and 71.9% accuracy.
• Preoperative Japanese Orthopaedic Association (JOA) score and imaging features were key predictors.

Conclusions:

• A predictive model using ML and deep learning for surgical outcomes in OPLL patients is feasible.
• These models show promise for application in spinal surgery for improved patient care.