Using Deep Learning for Accurate Recognition and Forecasting in the Automation of Diagnosing Diabetes

Abstract

Diabetes Mellitus (DM) is a global health problem with increasing incidence rates and more complications. There is nothing more critical than early diagnosis in the management of complications. The promising application of deep learning, a branch of artificial intelligence is in disease detection, and diagnosis because of its complexity in extracting the features from the data. This paper provides a critical analysis of applying deep learning techniques for early diagnosis of DM. To promote the timely diagnosis of DM, we first define the underlying disease process and its risk factors. Then, we go through the basics of deep learning and further gains some notion on architectures like CNNs, RNNs and their modifications with practical applications in DM detection. We talk about application of various types of data in deep learning such as Electronic Health Records, Medical Images and Wearables Data. Also, we discuss about the indicators used to assess the effectiveness of these models in respect to sensibility, selectivity, and area under curve of operating characteristic receiver. We also explain the issues that could arise when deploying deep learning models in clinical practices such as interpretation, expansion, as well as privacy issues. However, we also discussed emergent techniques like federated learning and the way in which transfer learning mitigates some of these issues. Furthermore, we also present some of the potential lines of work for future research such as the synthesis of multimodal data, the development of personalized prediction models for risk of DM along with the tracking of the status over time for timely intervention. In conclusion, this review minimize on the strengths of deep learning algorithm in early diagnosis of DM hence early intervention which is helpful in improving clients health status.