CNN-TumorNet: Leveraging Explainability in Deep Learning for Precise Brain Tumor Diagnosis on MRI Images

Novsheena Rasool ¹ Niyaz Ahmad Wani ^2* Javaid Iqbal Bhat ¹ Sandeep Saharan ³ Vishal Kumar Sharma ⁴ Bassma Saleh Alsulami ⁵ Hind Alsharif ⁶ Miltiadis D. Lytras ⁷

• ¹ Islamic University of Science and Technology, Awantipora, Jammu and Kashmir, India
• ² IILM University, Greater Noida, Uttar Pradesh, India
• ³ Bennett University, Greater Noida, Uttar Pradesh, India
• ⁴ Woxsen University, Hyderabad, Andhra Pradesh, India
• ⁵ King Abdulaziz University, Jeddah, Makkah, Saudi Arabia
• ⁶ Umm al-Qura University, Mecca, Saudi Arabia
• ⁷ American College of Greece, Athens, Attiki, Greece

Classifying brain tumors is crucial for early identification and precise diagnosis, facilitating timely treatment and improving patient outcomes. This research introduces CNN-TumorNet, a distinctive convolutional neural network (CNN) model utilizing deep learning to categorize MRI images into tumor or non-tumor classifications. CNN-TumorNet has demonstrated exceptional efficacy in brain tumor categorization, with an impressive 99% accuracy. Despite their formidable powers, deep learning models frequently encounter challenges regarding interpretability, which may restrict their widespread application. To address this, we employ the LIME (Local Interpretable Model-agnostic Explanations) methodology to elucidate the model's predictions, providing critical insights into its identification of malignant gliomas. The CNN-TumorNet excels in tumor classification inside MRI scans, owing to its exceptional accuracy and enhanced interpretability, rendering it an indispensable asset in medical diagnostics.