Densenet201:A Customized DNN Model for Multi-Class Classification and Detection of Tumors Based on Brain MRI Images
K. Sujatha, B. Srinivasa Rao
Abstract
In medical field, radiographic images are playing a vital role in detecting various diseases and injuries. Artificial Intelligence and machine learning techniques are very helpful to automate the process of examining these images in identifying the diseases. Brain tumor detection from Magnetic Resonance Images(MRI) can also be automated using deep learning models. Not only detecting the presence of a neoplasm in the brain images but also very important to recognize the type of neoplasm for proper treatment and to save the patient from life-threatening. This article aims at detecting the brain tumor type by customizing a pre-trained deep neural network DenseNet201. The brain tumor MRI images were collected from the bench mark datasets which contains 7023 MRI images of 4 classes ‘notumor’, ‘glioma’, ‘meningioma’, ‘pituitary’. The model was implemented by adjusting the final layer to predict four class labels present in the dataset. The developed model worked with an accuracy of 91% on training phase and with an accuracy of 88% on testing phase. This model can be enhanced by developing a hybrid model that combines deep learning algorithms used to extract features with ML algorithms used to classify different tumors based on the extracted features. The dataset considered for experimentation includes the brain MRI images related to only four classes, but there are nearly more than 150 different brain tumor types were documented in the research. Even though all 150 types of tumors may not be considered but it is very essential to consider at least some more tumor types that frequently occur in Indian children and adults for improving the automation process in detecting the type of tumor.