Deep Networks and Internet of Medical Things for Tracking the Post Surgical Recovery Condition: A Comparative Approach

Abstract

Proper post-operative recovery tracking is crucial for the early identification of complications and optimal patient outcomes. This paper introduces a computer-vision-based, sensor less technique fusing DL models with the Internet of Medical Things (IoMT) to predict a patient's recovery status Healthy or Unhealthy. Estimated vital signs such as heart rate (HR), blood pressure (BP), temperature, SpO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>, and hemoglobin are obtained from facial photographs and video inputs by applying machine learning algorithms. A synthetic dataset of 10,000 features is reduced in dimension to less than 2,000 dimensions by using Principal Component Analysis (PCA) and autoencoders. Three models are compared: a Baseline Deep Neural Network (DNN), Autoencoder + Multi-Layer Perceptron (MLP), and Residual MLP. Among these, Residual MLP yields the best classification accuracy of 94.7%, beating others because of improved gradient flow in deep architectures. A Streamlit-enabled web interface makes real-time vitals estimation and prediction possible. The findings highlight the viability of contactless, AI-based post-operative monitoring and stress the need for both feature compression and sound model selection. Real-world dataset expansion and mobile health deployment remain future directions.

Ähnliche Arbeiten