Abstract

The integration of Artificial Intelligence (AI) and the Internet of Medical Things (IoMT) presents significant computational and engineering challenges, especially in the deployment of real-time diagnostic systems. Therefore, this study proposed a numerically optimized and computationally efficient framework that combines deep learning (DL) architectures with IoMT-enabled deployment for the automated detection and classification of brain tumors using Magnetic Resonance Imaging (MRI). The methodology revolves around rigorous numerical preprocessing techniques, including normalized resizing, advanced data augmentation, and computationally efficient feature extraction via both custom and pre-trained Convolutional Neural Networks (CNNs). The key contributions of this study are tailored toward the evaluation of algorithmic performance beyond diagnostic accuracy, incorporating metrics such as model convergence, inference latency, memory footprint, and numerical stability under varied input conditions. The proposed AI-IoMT framework, known as I-BRAINDETECT, implemented as a web-based IoMT platform, demonstrates how algorithmic design and computational modeling can address the limitations of real-time medical image analysis. Performance evaluation and comparative analysis have shown that EfficientNetB0 and DenseNet121 achieved optimal performance in binary classification with 98.5 ± 0.202% accuracy, while ResNet50 excelled in multiclass classification with 95.4 ± 1.01% accuracy, both within a computationally constrained IoT environment. Validation of the trained models on an external dataset (Figshare) has shown that DenseNet121 achieved the best result with 93.99% accuracy. This work underscores the necessity of numerical robustness and algorithmic efficiency in bridging AI and IoT for scalable biomedical engineering solutions.