Improving Predictive Accuracy in Early Disease Detection Using Hybrid Neural Network Architectures and Feature Engineering
Authors: Sanjay Kumar Pandey, Dharmendra Kumar, Bechoo Lal
DOI: 10.87349/ahuri/181023
Page No: 35-61
Abstract
This research focuses on enhancing early disease detection through a novel hybrid neural network (HNN) framework, integrating Convolutional Neural Networks (CNNs) with Long Short-Term Memory (LSTM) networks. The HNN effectively handles diverse medical data, encompassing spatial information from medical imaging and temporal patterns from clinical time-series records, to improve diagnostic accuracy. The general methods of diagnosis have had problems related to low sensitivity and specificity, particularly when used to diagnose economic diseases at their initial stage. By combining the high performance associated with CNNs in image pattern recognition and the LSTMs in sequence modeling, the presented HNN model contributes to the improvement in the accuracy and interpretability of early identification of diseases such as cancer and cardiovascular diseases. Feature engineering is used to obtain specific characteristics from certain sources, including the texture and morphology of the medical image and the heart rate variability from the ECG data, enhancing the models. The HNN model was evaluated on two datasets: use of chest X-rays in the perception and diagnosis of diseases in the lungs as well as records of ECG in the perception and diagnosis of cardiovascular diseases. Performance is higher with the proposed hybrid model delivering a 94.2% accuracy score for X-ray image classification and 91.7% for ECG data classification compared to the individual models CNN, LSTM, DenseNet, ResNet, and Transformer. Therefore, this paper emphasizes the clinical applicability of the proposed hybrid deep learning models for early detection of diseases, highlighting that the incorporation of spatial as well as temporal data yields higher accuracy and sensitivity.
