Increasing radiologist workloads and increasing primary care radiology services make it relevant to explore the use of artificial intelligence (AI) and particularly deep learning to provide diagnostic assistance to radiologists and primary care physicians in improving the quality of patient care. This study investigates new model architectures and deep transfer learning to improve the performance in detecting abnormalities of upper extremities while training with limited data. DenseNet-169, DenseNet-201, and InceptionResNetV2 deep learning models were implemented and evaluated on the humerus and finger radiographs from MURA, a large public dataset of musculoskeletal radiographs. These architectures were selected because of their high recognition accuracy in a benchmark study. The DenseNet-201 and InceptionResNetV2 models, employing deep transfer learning to optimize training on limited data, detected abnormalities in the humerus radiographs with 95% CI accuracies of 83–92% and high sensitivities greater than 0.9, allowing for these models to serve as useful initial screening tools to prioritize studies for expedited review. The performance in the case of finger radiographs was not as promising, possibly due to the limitations of large inter-radiologist variation. It is suggested that the causes of this variation be further explored using machine learning approaches, which may lead to appropriate remediation.

Machine learning is often perceived as a sophisticated technology accessible only by highly trained experts. This prevents many physicians and biologists from using this tool in their research. The goal of this paper is to eliminate this out-dated perception. We argue that the recent development of auto machine learning techniques enables biomedical researchers to quickly build competitive machine learning classifiers without requiring in-depth knowledge about the underlying algorithms. We study the case of predicting the risk of cardiovascular diseases. To support our claim, we compare auto machine learning techniques against a graduate student using several important metrics, including the total amounts of time required for building machine learning models and the final classification accuracies on unseen test datasets. In particular, the graduate student manually builds multiple machine learning classifiers and tunes their parameters for one month using scikit-learn library, which is a popular machine learning library to obtain ones that perform best on two given, publicly available datasets. We run an auto machine learning library called auto-sklearn on the same datasets. Our experiments find that automatic machine learning takes 1 h to produce classifiers that perform better than the ones built by the graduate student in one month. More importantly, building this classifier only requires a few lines of standard code. Our findings are expected to change the way physicians see machine learning and encourage wide adoption of Artificial Intelligence (AI) techniques in clinical domains.

No abstract