A Cnn Approach for Accurate Stroke Diagnosis Using Brain Computed Tomography Imaging

Stroke detection from medical imaging, which is essential for early diagnosis and prompt intervention, greatly impacts patient outcomes. In this study, provide a technique that uses Convolutional Neural Networks (CNNs) to detect strokes from brain CT (computed tomography) pictures. A collection of 2,501 CT scans, including both normal and stroke cases, was used to develop and evaluate our model. A complete preprocessing pipeline was implemented, including training, validation, and systematic separation into test sets to enhance the data, scale the images, normalize, and reduce overfitting. The proposed CNN architecture consists of two convolutional blocks for feature extraction, followed by dense layers. According to the experimental results, the validation accuracy of the model was 92.93%, indicating its high capacity to distinguish between normal images and stroke. Additionally, the model's performance was thoroughly evaluated across a range of classes and thresholds using ROC curves, accuracy, recall, and F1-score. These comprehensive measurements demonstrated high sensitivity and specificity, underscoring the model's potential as a reliable clinical decision-support tool. According to the research, deep learning techniques - in particular, CNNs - have great potential for use in medical imaging applications for stroke diagnosis, as they can provide both high diagnostic accuracy and possible interpretability. Future studies will concentrate on growing the dataset and examining intricate structures and explanatory techniques in order to enhance therapeutic application.