A Deep Learning Approach Focusing on Diagnostic Sensitivity to Enhance Clinical Differentiation of Brain Tumors via Sequential ADC-MRI Analysis

Abstract

The differentiation of benign and malignant brain tumors using Apparent Diffusion Coefficient (ADC) MRI scans remains a challenge for clinicians, owing to the high variability of morphological features and the subtle signs of tissue densities. The present study proposes a simple yet highly effective deep learning-based framework for the classification of brain tumors as benign or malignant. The proposed framework incorporates a combination of Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM). The major advantage of the proposed framework is the use of seven consecutive image slices for brain tumor classification, unlike the conventional methods where a single individual image slice is considered. The use of seven consecutive image slices by the proposed framework actually attempts to capture the volumetric features of brain tumors, thus creating a more comprehensive picture of the brain tumor. The accuracy of the proposed framework for brain tumor classification is 98.05%, with a sensitivity of 100%, thus making the framework more reliable for the identification of malignant brain tumors and their safe elimination.