Artificial Intelligence Imaging Diagnosis Using Super-Resolution and 3D Shape for Lymph Node Metastasis of Low Rectal Cancer: A Pilot Study From a Single Center

Abstract

BACKGROUND: Although accurate preoperative diagnosis of lymph node metastasis is essential for optimizing treatment strategies for low rectal cancer, the accuracy of present diagnostic modalities has room for improvement. OBJECTIVE: To establish a high-precision diagnostic method for lymph node metastasis of low rectal cancer using artificial intelligence. DESIGN: A retrospective observational study. SETTINGS: A single cancer center and a college of engineering in Japan. PATIENTS: Patients with low rectal adenocarcinoma who underwent proctectomy, bilateral lateral pelvic lymph node dissection, and contrast-enhanced multi-detector row computed tomography (slice ≤1 mm) between July 2015 and August 2021 were included in the present study. All pelvic lymph nodes from the aortic bifurcation to the upper edge of the anal canal were extracted, regardless of whether within or beyond the total mesenteric excision area, and pathological diagnoses were annotated for training and validation. MAIN OUTCOME MEASURES: Sensitivity, specificity, positive predictive value, negative predictive value, and accuracy. RESULTS: A total of 596 pathologically negative and 43 positive nodes from 52 patients were extracted and annotated. Four diagnostic methods, with and without using super-resolution images and without using 3D shape data, were performed and compared. The super-resolution + 3D shape data method had the best diagnostic ability for the combination of sensitivity, negative predictive value, and accuracy (0.964, 0.966, and 0.968, respectively), while the super-resolution only method had the best diagnostic ability for the combination of specificity and positive predictive value (0.994 and 0.993, respectively). LIMITATIONS: Small number of patients at a single center and the lack of external validation. CONCLUSIONS: Our results enlightened the potential of artificial intelligence for the method to become another game changer in the diagnosis and treatment of low rectal cancer. See Video Abstract