AI-Enhanced Virtual Biopsies for Brain Tumor Diagnosis in Low Resource Settings

TL;DR

This work tackles the challenge of brain tumor diagnosis in low-resource settings by proposing a CPU-friendly 'virtual biopsy' pipeline that infers tumor characteristics from MRI. It combines a lightweight MobileNetV2 CNN with eight radiomics-style handcrafted features in a late fusion framework using a RandomForest classifier, together with Grad-CAM and feature-importance explanations for interpretability. Experiments on a four-class brain MRI dataset show that fusion outperforms single-branch models on validation, while revealing a notable dataset-shift gap at test time and sensitivity to reduced resolution and noise. The approach aims to provide non-invasive, decision-support tools suitable for resource-constrained environments, while acknowledging the need for broader validation, 3D modeling, and clinically approved radiomics pipelines before deployment.

Abstract

Timely brain tumor diagnosis remains challenging in low-resource clinical environments where expert neuroradiology interpretation, high-end MRI hardware, and invasive biopsy procedures may be limited. Although deep learning has achieved strong performance in brain tumor analysis, real-world adoption is constrained by computational demands, dataset shift across scanners, and limited interpretability. This paper presents a prototype virtual biopsy pipeline for four-class classification of 2D brain MRI images using a lightweight convolutional neural network (CNN) and complementary radiomics-style handcrafted features. A MobileNetV2-based CNN is trained for classification, while an interpretable radiomics branch extracts eight features capturing lesion shape, intensity statistics, and gray-level co-occurrence matrix (GLCM) texture descriptors. A late fusion strategy concatenates CNN embeddings with radiomics features and trains a RandomForest classifier on the fused representation. Explainability is provided via Grad-CAM visualizations and radiomics feature importance analysis. Experiments on a public Kaggle brain tumor MRI dataset show improved validation performance for fusion relative to single-branch baselines, while robustness tests under reduced resolution and additive noise highlight sensitivity relevant to low-resource imaging conditions. The system is framed as decision support and not a substitute for clinical diagnosis or histopathology.

Figures (10)

• Figure 1: Representative sample MRI images from the Kaggle dataset across the four classes: glioma tumor, meningioma tumor, pituitary tumor, and no tumor.
• Figure 2: Overall prototype pipeline for CNN--radiomics fusion-based "virtual biopsy" classification and explainability.
• Figure 3: CNN branch using MobileNetV2 for four-class tumor classification and for extracting 1280-dimensional embeddings for fusion.
• Figure 4: Radiomics feature importance from the RandomForest model, providing interpretable cues about which handcrafted descriptors contributed most to predictions.
• Figure 5: Hybrid fusion strategy concatenating deep embeddings and radiomics features, followed by a RandomForest classifier.