Design: Retrospective data from 2013 to 2022 was obtained for patients who underwent operative management for concern for torsion within a single healthcare system. This data was divided into training, validation, and test cohorts. Three algorithms were developed: (1)binary-classification decision tree; (2)simple neural network; (3)image-based neural network. The decision tree and simple neural network models included clinical and sonographic findings, and the image-based neural network model additionally included sonographic images. The primary outcome was performance in predicting torsion, as measured by area under the receiver operating characteristic curve (AUC).
Setting: N/A
Patients or Participants: Patients were included if they underwent surgery for possible torsion. Patients were excluded if they did not have sonographic images or an operative diagnosis. 477 patients were identified as possible candidates. 222 patients met inclusion criteria. Of these, 157 patients had torsion, and 65 did not have torsion.
Interventions: N/A
Measurements and Main Results: The models had test AUCs as follows: decision tree 0.5829; simple neural network 0.7572; image-based neural network 0.8053. In a high sensitivity test case, models had positive predictive values(PPV) and negative predictive values(NPV) as follows: decision tree PPV 75.8%/NPV 41.7%; simple neural network PPV 71.1%/NPV 0%; image-based neural network PPV 78.0%/NPV 100%.
Conclusion: This study demonstrates that the image-based neural network algorithm had improved detection of torsion compared to the decision tree or simple neural network models. Continued development of this image-based neural network algorithm could assist in clinical decision-making regarding adnexal torsion.
Lim, SL*1, Dong, H2, Darling, AJ3, Moyett, J4, Swartz, A3, Mazurowski, M2, Song, A3. 1Obstetrics & Gynecology, Duke University, Durham, NC; 2Department of Electrical & Computer Engineering, Duke University, Durham, NC; 3Obstetrics and Gynecology, Duke University, Durham, NC; 4Obstetrics and Gynecology, Columbia University, New York, NY