Design of a Secure and Accurate Deep Learning-Based Medical System for Stroke Detection

Abstract

Stroke is one of the most serious medical conditions that requires rapid and accurate diagnosis to avoid its fatal complications. The use of Deep Learning (DL) techniques contributes to enhancing the accuracy of detecting strokes in their early stages, enabling faster medical intervention and reducing long-term damage. In addition to early stroke detection, maintaining the confidentiality of patient information is vital to ensuring trust and protecting sensitive data. DL techniques also help detect unauthorized access or manipulation of patient data, enhancing digital security and ensuring compliance with healthcare privacy standards. In this paper, a precise and safe DL-based medical system for the detection of strokes is proposed. The design includes two stages. First, propose a high-accuracy DL learning algorithm to detect Stroke, which is achieved through a hybrid CNN-LSTM algorithm. The second stage is represented by maintaining the security of the proposed medical system by suggesting an efficient DL algorithm, which is attained using a lightweight Shallow Deep Neural Network (SDNN) algorithm. The performance evaluation of the two former stages, accuracy, security, and memory footprint, has been achieved. The simulation results demonstrate that the proposed hybrid CNN-LSTM algorithm achieves superior performance in stroke detection, reaching an accuracy of up to 99.6 percent when evaluated on the AHA stroke dataset. This performance surpasses that of standalone CNN and LSTM models. In terms of cybersecurity, the SzDNN algorithm also proves effective, maintaining a high detection rate of up to 96.8 percent using the NSL-KDD dataset. These findings highlight the model's strength in both accurate medical diagnostics and reliable network threat detection.