Cloud Based Optimal Brain Haemorrhage Classification using Ensemble Heun’s Convolutional Neural Networks to minimize the error rate

다국어 초록 (Multilingual Abstract)

In clinical diagnosis, accurate diagnosis of cerebral haemorrhage is important for timely intervention and enhanced patient outcomes. Because of its complexity and unpredictability, brain haemorrhage detection is still difficult, even with deep learning's promise in medical picture processing. The aim of this paper is to improve the performance of classification precision with rapid, fast, and low-cost services. Our proposed method focuses on detecting cloud-based brain haemorrhage disease and the classification of its subtypes using Ensemble Heun’s Convolutional Neural Networks and Enhanced Particle Swarm Optimization technique. The ensemble includes Global Average Pooling layers for effective training and a variety of deep learning models, including ResNet50, VGG16, and DenseNet121. In the existing blending ensemble technique, the various convolutional neural network models are used based on the low-order differential method. So, it should be some limitations, such as overfitting, bias, and security. Therefore, we have proposed a new ensemble Convolutional Neural Networks architecture designed by higher-order Heun’s numerical method to solve the neural ordinary differential equation for finding the better prediction and to reduce the overfitting problem. A useful collection of feature vectors is also derived by applying an improved particle swarm optimization technique. Moreover, the suggested work is implemented in a cloud-based framework to address the issues of data security, privacy preservation, performance enhancement, and quick diagnosis. According to the experimental data, the ensemble suggested method's optimal accuracy outperforms the current blending approach. Using a variety of performance criteria, including Accuracy, Precision, Recall, F1 Score, ROC-ACC, Error Rate, and testing prediction time of correctness, the suggested model produces outstanding classification results. When compared to the current approaches, our proposed approach performs well in the classification of brain haemorrhages. Finally, this research demonstrates the potential of optimized ensemble learning and cloud storage for reducing detection time in an emergency situation.