Integrating pyramid vision transformer and topological data analysis for brain tumor

Abstract

Read online

IntroductionBrain tumor (BT) classification is crucial yet challenging due to the complex and varied nature of these tumors. We present a novel approach combining a Pyramid Vision Transformer (PVT) with an adaptive deformable attention mechanism and Topological Data Analysis (TDA) to address the complexities of BT detection. While PVT and deformable attention have been explored in prior work, we introduce key innovations to enhance their performance for medical image analysis.MethodsWe developed an adaptive deformable attention mechanism that dynamically adjusts receptive fields based on tumor complexity, focusing on critical regions in MRI scans. The approach also incorporates an adaptive sampling rate with hierarchical dynamic position embeddings for context-aware multi-scale feature extraction. Feature channels are partitioned into specialized groups via an offset group mechanism to improve feature diversity, and a hierarchical deformable attention strategy further integrates local and global contexts to yield refined feature representations. Additionally, applying TDA to MRI images extracts meaningful topological patterns, followed by a Random Forest classifier for final BT classification.ResultsThe method was evaluated on the Figshare brain tumor MRI dataset. It achieved 99.2% accuracy, 99.35% recall, 98.9% precision, a 99.12% F1-score, a Matthews correlation coefficient (MCC) of 0.98, and a LogLoss of 0.05, with an average processing time of approximately 6 seconds per image.DiscussionThese results underscore the method's ability to combine detailed feature extraction with topological insights, significantly improving the accuracy and efficiency of BT classification. The proposed approach offers a promising tool for more reliable and rapid brain tumor diagnosis.

Keywords

• adaptive PVT
• BT classification
• brain tumor
• deep feature extraction
• Giotto-TDA
• Random Forest classifier