Applying Deep-Learning Algorithm Interpreting Kidney, Ureter, and Bladder (KUB) X-Rays to Detect Colon Cancer.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Oncology
• Radiology

Background:

• Early colorectal cancer (CRC) screening is vital for reducing mortality.
• Current methods like colonoscopy have low patient compliance and are invasive.
• Advanced imaging with AI shows promise but faces challenges in acquisition and cost.

Purpose of the Study:

• To develop a deep learning model (DLM) for detecting CRC using kidney, ureter, and bladder (KUB) radiographs.
• To evaluate the DLM's performance in detecting CRC on widely accessible KUB X-rays.

Main Methods:

• A retrospective study utilized KUB radiographs from two hospital cohorts (development, tuning, internal, and external validation sets).
• A 121-layer DenseNet convolutional network was trained for CRC detection.
• Model performance was assessed using receiver operating characteristic curves, sensitivity, specificity, and AUC.

Main Results:

• The DLM achieved an AUC of 0.738 (sensitivity 61.3%, specificity 74.4%) in internal validation and 0.656 (47.7%, 72.9%) in external validation.
• The model demonstrated better performance for high-grade CRC.
• AI-positive predictions correlated with increased long-term all-cause mortality risk.

Conclusions:

• AI-enhanced KUB X-ray analysis offers a potential cost-effective method to improve CRC screening coverage and effectiveness.
• This approach could serve as an alternative to traditional screening methods.
• Further prospective studies are needed for clinical integration.