Machine Learning-Aided Diagnosis Enhances Human Detection of Perilunate Dislocations
Anna Luan, MD, MS1,2, Lisa von Rabenau, MS1, Arman Serebrakian, MD2, Christopher S Crowe, MD3, Bao Do, MD1, Kyle R. Eberlin, MD4, James Chang, MD5 and Brian Pridgen, MD6, (1)Stanford University, Palo Alto, CA, (2)Massachusetts General Hospital, Boston, MA, (3)University of Washington, Seattle, WA, (4)Massachusetts General Hospital/Harvard Medical School, Boston, MA, (5)Plastic & Reconstructive Surgery, Stanford University, Stanford, CA, (6)The Buncke Clinic, San Francisco, CA
Introduction: Perilunate injuries can cause significant morbidity when untreated. Unfortunately, perilunate/lunate dislocations are frequently misdiagnosed or missed altogether, with approximately 25% of cases missed at the initial presentation. Initial interpretation of radiographs may often be performed by trainees or clinicians without subspecialty training. New artificial intelligence technologies have the potential to support physicians’ clinical judgement, but applications within hand surgery have thus far been relatively limited. We hypothesize that utilization of a machine learning algorithm can improve human detection of perilunate/lunate dislocations.
Methods: Performance of a machine learning algorithm was validated. Human participants from emergency medicine, hand surgery, and radiology were asked to evaluate for presence of a perilunate/lunate dislocation on 30 distinct lateral wrist radiographs with and without the use of a machine learning algorithm, which was used to label the lunate. Human performance with and without the machine learning tool was evaluated using accuracy, sensitivity, specificity, and total response time.
Results: A total of 137 participants were recruited, with 55 respondents from emergency medicine, 33 from radiology, and 49 from hand surgery. Thirty-nine participants were attending physicians or fellows and 98 were residents. Use of the machine learning tool improved accuracy from 88.7% to 92.5%, specificity from 87.5% to 94.4%, and resulted in a 28.5% reduction in average diagnosis time. When stratified by training level, attending physicians and fellows had an improvement in specificity from 92.5% to 97.1% and a 29.9% reduction in time to diagnosis, without improvement in accuracy or sensitivity. For residents, use of the machine learning tool resulted in improved accuracy from 86.1% to 91.0%, specificity from 85.6% to 93.3%, and a 27.9% reduction in diagnosis time. Finally, performance of surgery and radiology residents when assisted by the machine learning tool improved to mimic performance of attendings and fellows in accuracy and specificity.
Conclusions: Use of a machine learning tool improves resident accuracy for radiographic detection of perilunate dislocations, and specificity and time to diagnosis for all training levels. This machine learning tool may help to decrease misdiagnosis of perilunate dislocations, particularly when subspecialist evaluation is delayed, and to decrease the time required for diagnosis.
Stratified Outcomes.jpg
On ROC Curve.jpg
Back to 2024 Abstracts