MobileNet and Streamlit Applications for Classification and Prediction of MRI-Based Types of Brain Tumor

MobileNet and Streamlit Applications for Classification and Prediction of MRI-Based Types of Brain Tumor

Publication Date : 2024-05-15
Author(s) :

G. Kharmega Sundararaj

           
Article Name :

MobileNet and Streamlit Applications for Classification and Prediction of MRI-Based Types of Brain Tumor

Abstract :

The rapid proliferation of abnormal cells within the cerebrum and its surrounding areas leads to the formation of tumor cells with intricate and challenging patterns. Consequently, relying solely on magnetic resonance imaging (MRI) for brain tumor detection presents complexities due to the need for precise identification amidst normal anatomical structures and potentially anomalous tissues. Recognizing the urgency of early and precise diagnosis in combating brain cancers, this paper proposes a methodology for classifying and predicting MRI based brain tumor types using MobileNet through a Streamlit application. Initially, the input image dataset undergoes preprocessing to enhance image quality, facilitated by a Gaussian filter. Subsequently, the FCM segmentation technique is employed to accurately delineate tumors’ size, location, and morphology. Finally, feature extraction and classification are performed utilizing the MobileNet framework, a deep learning method known for its accuracy and efficiency, particularly in medical image segmentation. Performance evaluation metrics such as precision, recall, and accuracy are computed, culminating in the calculation of the F score to assess the model’s effectiveness. This work leverages the Streamlit environment, a Python platform for developing web based interactive applications, to facilitate seamless user interaction. Ultimately, the predicted output images depicting various brain tumor types glioma, meningioma, pituitary, and absence of tumor are displayed within the Streamlit software interface. By offering a swift and precise means of disease classification, this approach aids healthcare professionals in promptly identifying tumor afflicted patients while streamlining computational overhead. The proposed classifier shows the highest specificity of 94.5 % and sensitivity of 97.6%.

No. of Downloads :

17