الكشف والتصنيف الآلي لورم الدماغ البشري بأستخدام التعلم الآلي

AUTOMATED DETECTION AND CLASSIFICATION OF HUMAN BRAIN TUMOR USING MACHINE LEARNING

Abstract

The automated system is important for helping doctors and radiologists to detect and classify brain tumors. Solid tumors inside the skull result from uncontrolled and abnormal cell division. The main challenge in detecting brain tumors is the difference in tumor location, shape, size, and the diversity and complexity of brain tumors. Machine learning and Deep learning are the perfect solution to this problem.

The proposed work includes data pre-processing and image segmentation, two segmentation techniques (edge-based segmentation and region-based segmentation) were applied to compare them, where the results using region were better than the results using edge. Morphological operations was applied after the segmentation process and includes the closing processes(dilation, and erosion).

Next, machine learning and deep learning algorithms were applied to classify MRI brain tumors into four types: pituitary, glioma, meningioma, and no tumor in normal cases, based on a specific set of features that improve classification accuracy, save time and cost.

In this work, two proposed models were implemented. The first is the combination Visual Geometry Group 16 (VGG16) with four traditional classifiers: Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and Naive Bayes (NB). The combination is implemented due to its deep learning capabilities, as it can extract complex features such as details of brain tumors. The second proposed model is a convolutional neural network (CNN).

In this work, experimental results showed that the combination ((VGG16)with random forest) using region-based segmentation obtained the accuracy of 99.24%. The percentage is higher compared to the combination ((VGG16)with random forest) when using edge-based segmentation where was the result 98.78%.

This dataset is originally a combination of three datasets: Figshare, the SARTAJ data set, and the Br35H dataset, containing MRI images of the four types of brain tumors was used.

Finally, our results were compared with existing research in the field of segmentation and classification on the same dataset, where our results proved to be the best. Our proposed model achieved an accuracy of 99.24%, while previous research results ranged from 97.12%, 95.73%, to 87.67%.