Detection of Intracranial Hemorrhage - A Comparative Study of Traditional Machine Learning Techniques and Mask Generation using CNN

Intracranial Hemorrhage (ICH) is a life-threatening condition that occurs due to bleeding within the human brain. Medical imaging techniques such as Angiogram, CT angiography, Computed Tomography (CT), Lumbar Puncture, and Magnetic Resonance Imaging (MRI), combined with cutting-edge technology, machine learning techniques, and deep learning algorithms have proved to be efficient in the classification and detection of intracranial hemorrhage. This paper provides a comparative study and extensively carries out brain hemorrhage classification using different machine learning models such as kNN, SVM, Decision tree (AdaBoost and Gradient Boosting), Random Forest and Naïve Bayes. The convolutional neural network technique such as AlexNet employed for classification gives better results. Intracranial hemorrhage masking using CNN helps us to perceive the region of hemorrhage using DenseNet-121 and is successful in visualizing the source of bleeding. The open-source Kaggle platform is used for dataset acquisition and comparative study. The results revealed that Gradient Boosting proffers a maximum accuracy of 96.7% whereas the models such as kNN, SVM, Random Forest and Naïve Bayes have achieved an accuracy of 94.5%, 91.8%, 94.3% and 78.8% respectively.