A Web-Deployed, Explainable AI System for Comprehensive Brain Tumor Diagnosis.

Area of Science:

• Neuro-oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate brain tumor diagnosis is crucial for treatment planning in neuro-oncology.
• Deep learning models for 2D classification and 3D segmentation can enhance radiological workflows.
• Explainable AI (XAI) techniques improve the interpretability of these models.

Purpose of the Study:

• To develop a web-based platform for brain tumor segmentation and classification diagnosis.
• To integrate 2D classification and 3D segmentation using deep learning.
• To incorporate explainable AI for enhanced model interpretability.

Main Methods:

• A diagnosis system combining 2D tumor classification (MobileNetV2) and 3D volumetric segmentation (SegResNet) was developed.
• A meta-classifier MLP was used for binary tumor detection.
• Explainability was provided using XRAI maps and Gaussian overlays, integrated into a web interface.

Main Results:

• The 2D MobileNetV2 model achieved 98.09% accuracy for tumor classification.
• The 3D SegResNet model obtained Dice coefficients of 68-70% for tumor segmentations.
• The MLP tumor detection module achieved 100% accuracy, with explainability modules consistent with pathological features.

Conclusions:

• The deep learning diagnosis system improves brain tumor classification and segmentation with interpretable outcomes.
• The web-based tool with a user-friendly interface is suitable for clinical radiology workflows.
• Explainable AI techniques enhance the clinical utility of deep learning in neuro-oncology.