Enabling earlier detection of spinal lesions in CT imaging with artificial intelligence-a case study

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Spinal metastases are common in advanced cancer, impacting patient staging and treatment.
• Delayed or missed detection of malignant spinal lesions can negatively affect patient outcomes.

Purpose of the Study:

• To evaluate if an AI-based second reader can detect malignant spinal lesions on computed tomography (CT) scans earlier than radiologists.
• To assess the potential of AI in supporting precision oncology through timely diagnosis and treatment planning.

Main Methods:

• A 3D nnU-Net segmentation model was trained on 653 CT scans.
• The model was applied to a retrospective cohort of 200 patients with confirmed malignant spinal lesions.
• Radiologists reviewed scans unaided and with AI support to compare detection rates and lead times.

Main Results:

• The AI system identified 12 malignant spinal lesions missed by radiologists during unaided review, with a mean lead time of 228 days.
• AI highlighted an additional 25 retrospectively visible but initially unreported malignant lesions.
• AI-assisted review led to earlier flagging of findings in 37 patients.

Conclusions:

• AI-assisted CT interpretation shows potential for identifying subtle or overlooked malignant spinal lesions earlier than standard reporting.
• These findings suggest AI may enhance diagnostic completeness and facilitate earlier treatment initiation.
• Further prospective studies are needed to confirm clinical benefits and impact on patient outcomes.