The Diagnosis of Developmental Dysplasia of the Hip From Hip Ultrasonography Images With Deep Learning Methods.

Autor:	Atalar H; Department of Orthopaedics and Traumatology, Faculty of Medicine, Gazi University., Üreten K; Department of Rheumatology, Faculty of Medicine, Ufuk University.; Department of Computer Engineering., Tokdemir G; Department of Computer Engineering, Faculty of Engineering, Çankaya University., Tolunay T; Department of Orthopaedics and Traumatology, Faculty of Medicine, Gazi University., Çiçeklidağ M; Department of Orthopaedics and Traumatology, Faculty of Medicine, Gazi University., Atik OŞ; Turkish Joint Diseases Foundation, Çankaya, Ankara, Turkey.
Jazyk:	angličtina
Zdroj:	Journal of pediatric orthopedics [J Pediatr Orthop] 2023 Feb 01; Vol. 43 (2), pp. e132-e137. Date of Electronic Publication: 2022 Nov 04.
DOI:	10.1097/BPO.0000000000002294
Abstrakt:	Background: Hip ultrasonography is very important in the early diagnosis of developmental dysplasia of the hip. The application of deep learning-based medical image analysis to computer-aided diagnosis has the potential to provide decision-making support to clinicians and improve the accuracy and efficiency of various diagnostic and treatment processes. This has encouraged new research and development efforts in computer-aided diagnosis. The aim of this study was to evaluate hip sonograms using computer-assisted deep-learning methods. Methods: The study included 376 sonograms evaluated as normal according to the Graf method, 541 images with dysplasia and 365 images with incorrect probe position. To classify the developmental hip dysplasia ultrasound images, transfer learning was applied with pretrained VGG-16, ResNet-101, MobileNetV2 and GoogLeNet networks. The performances of the networks were evaluated with the performance parameters of accuracy, sensitivity, specificity, precision, F1 score, and AUC (area under the ROC curve). Results: The accuracy, sensitivity, specificity, precision, F1 score, and AUC results obtained by testing the VGG-16, ResNet-101, MobileNetV2, and GoogLeNet models showed performance >80%. With the pretrained VGG-19 model, 93%, 93.5%, 96.7%, 92.3%, 92.6%, and 0.99 accuracy, sensitivity, specificity, precision, F1 score, and AUC results were obtained, respectively. Conclusion: In this study, in addition to the ultrasonography images of dysplastic and healthy hips, images were also included of probe malpositioning, and these images were able to be successfully evaluated with deep learning methods. On the sonograms, which provided criteria appropriate for evaluation, successful differentiation could be made of healthy hips and dysplastic hips. Level of Evidence: Level-IV; diagnostic studies. Competing Interests: The authors declare no conflicts of interest. (Copyright © 2022 Wolters Kluwer Health, Inc. All rights reserved.)
Databáze:	MEDLINE
Externí odkaz:	Zobrazit plný text záznamu