Title: Integrating Topology with Deep Representation Learning: A TDA-ViT Hybrid Approach for Four-Class Brain Tumor Diagnosis

Abstract:

Precise classification of brain tumors using magnetic resonance imaging (MRI) is essential for facilitating early diagnosis and guiding clinical decisions. While Vision Transformers (ViTs) have demonstrated remarkable efficacy in medical image analysis by capturing global contextual features, they frequently struggle to identify the inherent structural and topological patterns within tumor areas. To overcome this challenge, we introduce a hybrid framework that merges features derived from Topological Data Analysis (TDA) with representations learned by pretrained Vision Transformers to classify brain tumors into four distinct categories.

In this methodology, TDA is employed to generate complementary topological descriptors that elucidate the geometric structure, connectivity, and shape characteristics embedded in MRI scans. Simultaneously, a pretrained ViT model extracts high-level semantic information from the same imagery. These two distinct feature spaces are subsequently integrated to create a cohesive, more discriminative representation suitable for classification tasks.

The proposed model was assessed using the BRISC2025 dataset, which comprises four classes: glioma, meningioma, pituitary tumors, and non-tumor cases. Our findings indicate that the synergy of topological and transformer-based features yields superior performance compared to relying on either method in isolation. The TDA-ViT fusion model recorded an accuracy of 99.10%, a precision of 99.27%, a recall of 99.15%, an F1-score of 99.21%, and an AUC of 99.98%. Furthermore, it surpassed the performance of several leading state-of-the-art architectures, such as ResNet50, ResNet101, EfficientNetB2, and standalone Vision Transformers. These outcomes highlight that topological features offer significant complementary value to deep representation learning, resulting in a robust and highly accurate system for the automated classification of brain tumors.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC