Computational methods for AI-based healthcare Engineering
Main Article Content
Abstract
The problem of Diabetes Mellitus (DM) is a current and acute one, the incidence increases and tends to involve more and more people worldwide. It also is relevant with regard to management and prevention of complications, as the skills of early detection are crucial. Computer assisted diagnosis using artificial intelligence such as deep learning has been identified to have potential application in diagnosing diseases because of its feature of identifying complex relationships from a large amount of data. This paper provides an excellent review of employing DL for early diagnosis of DM as discussed next. In our first focused topic, we explain the condition of DM in relation to its causes and contributors to the disease’s progression, as well as emphasizing the significance of early screening. Then, we go through key architectures in deep learning, including convolution neural networks (CNNs), recurrent neural networks (RNNs) and their derivatives which have been utilized in the study of DM detection. We consider the application of different types of data such as EHR, MRI, and data from wearable sensors in deep learning. In addition, we discuss the factors used to assess the performance of these models including sensitivity, specificity, and area under the curve of ROC. We also discuss the related issues of the clinical application of deep learning, including explanation, extension and data protection. However, some of the challenges include subgroup imbalances, overlapping data, and privacy violations, all of which we explain further with recent innovations in the field; federated learning and transfer learning. Furthermore, we describe potential research directions such as multi-source data fusion, individualized risk assessment, and the integration of continuous tracking for DM’s early intervention. More specifically and in turn, this review also points to the high possibility of deep learning algorithms in the early diagnosis of DM to enhance optimum management.