Evaluating Performance of a Deep Learning-based Artificial Intelligence Model for Ovarian Tumor Classification Using a Multicenter CT Dataset

Purpose: To develop a deep learning–based, computer-aided diagnosis (CADx) model for preoperative classification of ovarian tumors (OTs) on CT scans and to compare its performance with current US models and radiologist assessments. Materials and Methods: This retrospective multicenter study (January 2021–November 2023) included patients with indeterminate OTs. The dataset comprised training, internal (n = 360), and external test (n = 27) sets. Final histopathology served as the reference standard. The CADx model was trained using self-supervised learning on public and institutional CT datasets. Performance of the CADx model was compared with that of two current US-based models (Risk of Malignancy [RMI] and Assessment of Different NEoplasias in the adneXa [ADNEX] models) and with radiologist reports. Metrics included the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and positive and negative predictive values, with comparisons assessed using 95% CI overlap. Results: The dataset contained 387 OT images from 344 patients (226 benign and 118 malignant OTs). The model achieved a median AUC of 0.84 (95% CI: 0.65, 0.92) on the internal test set and 0.61 (95% CI: 0.59, 0.65) on the external test set. The CADx model performed comparably with the two US models and radiologists. On the internal test set, AUCs for RMI, ADNEX, and radiologists were 0.77 (95% CI: 0.72, 0.83), 0.68 (95% CI: 0.51, 0.84), and 0.76 (95% CI: 0.69, 0.83), respectively. On the external test set, corresponding AUCs were 0.66 (95% CI: 0.44, 0.88), 0.86 (95% CI: 0.60, >0.99), and 0.67 (95% CI: 0.26, >0.99), respectively. The CADx model yielded the highest sensitivity (94.7%). Conclusion: Despite disease and data variability, this CT-based deep learning model for preoperative OT classification achieved comparable performances to US models and radiologists on internal and external test sets, but further refinement is needed before clinical implementation. Clinical trial registration no. NTC05174377.