Deep learning assessment compared to radiologist reporting for metastatic spinal cord compression on CT

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Metastatic spinal cord compression (MSCC) is a severe complication of advanced cancer.
• Timely diagnosis of MSCC is crucial for effective management and patient outcomes.
• Computed tomography (CT) is a key imaging modality for diagnosing MSCC.

Purpose of the Study:

• To externally validate a deep learning (DL) algorithm for classifying MSCC on CT scans.
• To compare the performance of the DL algorithm against radiologist assessments.
• To evaluate the potential of DL in expediting MSCC diagnosis.

Main Methods:

• Retrospective collection of CT scans from patients with suspected MSCC.
• Development and internal testing of a DL algorithm for MSCC classification.
• External validation of the DL algorithm using a separate test set.
• Comparison of DL algorithm performance with assessments by multiple radiologists and CT reports.

Main Results:

• The DL algorithm achieved high inter-rater agreement for MSCC grading (kappa=0.872 internally, 0.844 externally).
• DL algorithm performance was superior to individual radiologists on internal and external testing.
• CT report classification showed poor agreement and low sensitivity for high-grade MSCC compared to the DL algorithm.

Conclusions:

• The DL algorithm for MSCC classification on CT demonstrates high accuracy and reliability.
• The DL algorithm outperforms radiologist assessments and CT reports in identifying MSCC.
• This AI tool has the potential to significantly aid in the earlier diagnosis of MSCC.