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

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Ovarian tumors (OTs) require accurate preoperative classification for optimal patient management.
• Current diagnostic models, including US-based tools, have limitations in classifying indeterminate OTs.
• Deep learning offers potential for improving diagnostic accuracy in medical imaging.

Purpose of the Study:

• To develop and evaluate a deep learning-based computer-aided diagnosis (CADx) model for preoperative classification of ovarian tumors using CT scans.
• To compare the performance of this novel CADx model against established US-based models (RMI and ADNEX) and radiologist assessments.

Main Methods:

• A retrospective multicenter study included 387 indeterminate ovarian tumors from 344 patients.
• A deep learning CADx model was trained using self-supervised learning on diverse CT datasets.
• Performance was assessed using AUC, sensitivity, specificity, and predictive values, comparing the CADx model with RMI, ADNEX, and radiologist reports on internal and external test sets.

Main Results:

• The deep learning CADx model demonstrated comparable performance to US models and radiologists across internal and external test sets.
• The model achieved a median AUC of 0.84 (internal) and 0.61 (external).
• The CADx model achieved the highest sensitivity at 94.7%.

Conclusions:

• The CT-based deep learning CADx model shows promise for preoperative ovarian tumor classification, achieving performance comparable to existing methods.
• Despite variability in disease and data, the model's results suggest its potential utility.
• Further research and refinement are necessary before widespread clinical implementation.