DOI: 10.46810/tdfd.1749282 ISSN: 2149-6366

Enhancing Reliability in Deep Learning Diagnosis of Brain Tumors Using Grad-CAM

Musab Coskun
Show PDF Cite
While deep learning models in medical imaging have gained popularity as a means to enhance patient outcomes and diagnostic accuracy recently, one of their main issues is interpretability, which is essential to understanding and debugging the model. Explainable Artificial Intelligence (XAI) is a recent rising research direction that aims to explain this black box part of the deep learning models. For quick identification, clinical evaluations and imaging methods such as Magnetic Resonance Imaging (MRI) scans, are frequently utilized; nevertheless, manual analysis has challenges such as subjectivity and delays. On the other hand, AI-based models convey a more rapid and reliable approach to classifying and detecting brain tumors. In this work, we present a transparent and explainable framework of pre-trained Convolutional Neural Network (CNN) models combined with Gradient Weighted Class Activation Mapping (Grad-CAM) for the classification of brain MRI images. We compare the effectiveness of ResNet50, DenseNet121, MobileNetV2 and ConvNeXtTiny architectures. We obtained a test accuracy of 100% and precision-recall scores above 99.90%, highlighting the model's effectiveness in identifying whether a tumor is present. The results illustrate how the models have enhanced localization skills by visualizing the regions of focus in the predictions through the application of the Grad-CAM method. This blend of interpretability offers a promising step toward creating more reliable and understandable tools for diagnosing brain tumors.

More from our Archive