Optimizing Telemedicine Flow Across Imbalanced Multiple Classes with different Arabic Dialects

Abstract

This compendium focuses on various aspects of improving medical diagnosis, recommendation generation, symptom identification, quality assessment of telemedicine consultations, and specialty detection in the Arabic language. These studies aim to enhance healthcare services and patient-doctor interactions in the context of telemedicine. The first set of papers addresses the development of intelligent systems for medical diagnosis and decision support. These systems utilize machine learning algorithms trained on large datasets of patient questions and structured symptoms where all data sets obtained from Altibbi Telemedicine databases. By combining different modalities and employing various feature representation techniques and classifiers, these systems demonstrate promising predictive abilities and accuracy in predicting patient conditions. Another set of papers focuses on natural language processing (NLP) applications in telemedicine, particularly in generating medical recommendations and analyzing healthcare-related text. Deep learning-based models are developed to simplify the process of writing medical recommendations in Arabic. These models achieve impressive results in next word prediction and show potential for improving service satisfaction and patient-doctor interactions. Additionally, one of the researches has been conducted on word embedding models specifically designed for medical and healthcare applications in the Arabic language. By training neural-based word embedding models on large datasets of medical consultations and questions, these studies demonstrate the effectiveness of Word2Vec and fastText models in capturing the semantics of text, thereby improving the performance of healthcare NLP-based applications. Furthermore, deep learning approaches are explored for automated question classification and symptom identification from unstructured medical consultations. By utilizing deep neural networks and domain-specific word embedding models, these studies achieve high accuracy rates in classifying medical questions into medical specialities and identifying symptoms from Arabic texts, thereby assisting doctors in the diagnosis process and improving the efficiency and accuracy of telemedicine consultations. Lastly, the research delves into the automation of quality assessment in voice-based telemedicine consultations and specialty detection in highly imbalanced multiclass datasets. Deep learning models, combined with statistical and spectral information, are developed to assess the quality of patient-doctor conversations and detect the correct medical specialty for each question. Various techniques, such as oversampling and keyword identification, are employed to improve the performance of specialty detection, which has implications for customizing consultation flows and minimizing the doctor’s effort in addressing the correct specialty. Overall, these research papers contribute to advancing telemedicine services in the Arabic context by leveraging machine learning, deep learning, and NLP techniques to enhance medical diagnosis, recommendation generation, symptom identification, quality assessment, and specialty detection.