CT-based clinical-radiomics model for preoperative identification of appendiceal neoplasms in patients undergoing surgery for suspected appendicitis

Background/objective Appendiceal neoplasms may mimic acute appendicitis and can be overlooked on routine CT. Failure to recognize a tumor before surgery may affect operative planning, intraoperative precautions, specimen handling, and subsequent management. We developed and internally validated a CT-based clinical-radiomics model in patients undergoing surgery for suspected appendicitis and assessed whether it could identify pathologically confirmed neoplasms that had not been suspected in the original CT report. Methods This retrospective, single-center study included 520 patients who underwent surgery for suspected appendicitis and had pathologic confirmation: 458 had appendicitis and 62 had appendiceal neoplasms. Clinical, radiomics, and combined models were evaluated using patient-level stratified five-fold cross-validation, with all preprocessing confined to the training data in each split. We reviewed the original CT report for each patient with a neoplasm to determine whether a tumor had been suspected before surgery. At a threshold selected using a prespecified sensitivity-oriented rule, we quantified additional tumor detection and the resulting number of model-positive patients. Results Of the 62 appendiceal neoplasms, 52 (83.9%) were low-grade appendiceal mucinous neoplasms. The original CT report raised suspicion of a neoplasm in 22 cases (35.5%). The combined model identified 53 neoplasms (85.5%), including 32 of the 40 tumors that had not been suspected in the original report (80.0%). At this threshold, 76 of 520 patients (14.6%) were model-positive, including 23 false-positive cases. Each additional tumor identified beyond the original report therefore corresponded to roughly 2 to 3 model-positive cases. The pooled out-of-fold AUC was 0.976 (95% CI 0.960-0.988), and the PR-AUC was 0.875 (95% CI 0.793-0.935). Conclusion In this surgical cohort, the combined clinical-radiomics model identified additional pathologically confirmed appendiceal neoplasms that had not been suspected in the original CT report. The model may support routine CT interpretation by flagging patients who warrant closer preoperative review. External validation and reader-aided evaluation are needed before clinical implementation.