Classification of Musculoskeletal Abnormalities with Convolutional Neural Networks

• Guilherme Tiaki Sassai Sato UNICAMP http://orcid.org/0000-0002-6615-7268
• Leodécio Braz da Silva Segundo UNICAMP http://orcid.org/0000-0002-9326-4122
• Zanoni Dias UNICAMP http://orcid.org/0000-0003-3333-6822

Resumo

Computer-aided diagnosis has the potential to alleviate the burden on medical doctors and decrease misdiagnosis, but building a successful method for automatic classification is challenging due to insufficient labeled data. In this work, we investigate the usage of convolutional neural networks to diagnose musculoskeletal abnormalities using radiographs (X-rays) of the upper limb and measure the impact of several techniques in our model. We achieved the best results by utilizing an ensemble model that employs a support vector machine to combine different models, resulting in an overall AUC ROC of 0.8791 and Kappa of 0.6724 when evaluated using an independent test set.

Palavras-chave: Deep learning, Musculoskeletal abnormalities, X-ray