Multimodal data analysis for post-decortication therapy optimization using IoMT and reinforcement learning

Abstract Multimodal neural network approaches mirror clinical decision-making processes for therapy optimization. Multiple data sources, including images, electronic health records, and the Internet of medical things, can be challenging for precise diagnoses. In this research, reinforcement learning, specifically deep Q-learning (DQL), utilizing multimodal data, has been employed to determine the most suitable treatment plans, particularly for patients undergoing lung decortication surgery. The model performance has been evaluated using rewards, epsilon decay, and Q -values across three different actions. The model’s performance has also been compared with machine learning models, such as Naïve Bayes, K-nearest neighbor, random forest, logistic regression, and support vector machine, regarding several performance metrics, including accuracy, precision, recall, and the area under the curve. Our findings demonstrate that the DQL model effectively learns optimal actions, significantly enhancing therapy optimization.