Title : AI guided navigation of high risk incidental findings on imaging studies leads to cancer diagnoses
Abstract:
Introduction: Artificial Intelligence (AI) is a burgeoning tool in the field of surgery. AI technology is vastly unexplored with few applications currently in use. High risk incidental findings in radiologic studies are recognized as a critical source of missed or delayed cancer diagnoses due to gaps in communication, manual tracking burdens, and limited resources. AI has the potential to fundamentally improve recognition of high risk incidental findings on imaging studies and aid with navigation to timely oncologic care.
Methods: We utilized a commercial end-to-end oncology focused natural language processing (NLP) solution that employed machine learning to identify and track high risk incidental findings in real-time. The software reviewed imaging studies performed at emergency and outpatient imaging facilities between 6/1/2024 and 6/30/2025. The software model was provided with keywords and parameters to identify high risk lung, liver and pancreas lesions for malignancy. The software identified hits among the reports and then a High Risk Incidental Findings Team(HRIFT) manually reviewed the radiology reports for validation. Patients with new high risk findings without established oncologic care were navigated.
Results: Over 13 months, 961,298 imaging studies were reviewed by the software and 48,605 (5%) reports had high risk lesions identified. The HRIFT reviewed all 48,605 imaging reports and confirmed 43,708 clinically significant high risk findings. AI software demonstrated a rate of 4.5% high risk incidental findings among all scans reviewed, and a 0.5% false positive rate. Among the confirmed high risk incidental findings group, 4,833 patients did not have an established specialty provider. 416 of these patients were successfully navigated for multidisciplinary care. After additional work up, 59 patients were ultimately diagnosed with cancer, including 13 (22%) liver, 39 (66%) lung and 7 (12%) pancreatic cancers. This demonstrates an incidence of 0.1% new cancer diagnoses among all validated reports and 14% incidence among navigated patients.
Conclusion: AI technology identified high risk lesions in approximately 5% of patients reviewed, which would have taken significantly more time to be worked up by traditional means or might have been missed altogether. The impact on patients may include decreased time to diagnosis and increased rates of follow up with specialty care. The machine learning algorithms and validation will enable improved model performance by modifying the variables entered. This project highlights that AI technology has vast potential for streamlined patient care and helps support multidisciplinary teams in ensuring that actionable findings are appropriately identified and managed. AI technology may fundamentally change the ability of patients to receive prompt and efficacious cancer care.
