Abstract

The application of machine learning approaches to diagnose gliomas from magnetic resonance image data is becoming increasingly common as machine learning algorithms are being refined and inexpensive, high power computing resources are readily available. Most current glioma detection techniques involve a biopsy which is invasive and often has several long-lasting negative impacts. Many different classification techniques have been tried for image classification of scans containing gliomas including traditional models such as K-nearest neighbors and Random Forest. Deep learning models, specifically convolutional neural networks and pre-built models, have also been applied to the situation. This study applied 2 traditional machine learning models, Support Vector Machine and logistic regression classifier, to a publicly available dataset containing over 1200 brain scans of healthy and diseased patients. After normalization, 854 images were used to train the two models and 367 images were used to test the models. They were then evaluated on the parameters of: area under the receiver operating characteristic curve and sensitivity to determine which model performed better. Both models had similar AUC scores, but the SVM model had much higher sensitivity. Considering the possibly fatal ramifications of incorrectly diagnosing a patient who has been infected with glioma, it was determined that the SVM was better equipped to handle the classification task. Computer aided diagnosis of the tumor will hopefully be able to increase the survival rate of people diagnosing with gliomas.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call