Deep Multi-Instance Conv-Transformer Frameworks for Landmark-Based Brain MRI Classification

• DOI: 10.3390/electronics13050980
• Publication Date: 2024-03-04T15:11:57Z
• Authors: Guannan Li; Zexuan Ji; Quansen Sun
• Keywords: Landmark

For brain diseases, e.g., autism spectrum disorder (ASD), with unclear biological characteristics, the detection of imaging-based biomarkers is a critical task for diagnosis. Several landmark-based categorization approaches have been developed computer-aided diagnosis such as Alzheimer’s disease (AD), utilizing structural magnetic resonance imaging (sMRI). With automatic landmarks disease, more detailed features were identified clinical Multi-instance learning an effective technique classifying diseases based on landmarks. The multiple-instance approach relies assumption independent distribution hypotheses and mostly focused local information, thus correlation among different regions may be ignored. However, according to previous research ASD AD, abnormal development highly correlated. Vision Transformers, self-attention modules capture relationship between embedded patches from whole image, recently demonstrated superior performances in many computer vision tasks. Nevertheless, utilization 3D MRIs imposes substantial computational load, especially while training Transformer. To address challenges mentioned above, this research, we proposed multi-instance Conv-Transformer (LD-MILCT) framework solution aforementioned issues In network, two-stage strategy was explore both spatial morphological information regions; Transformer utilizes head (MIL head) fully utilize that are not involved ultimate classification. We assessed our using T1-weighted MRI images AD databases. Our method outperformed existing deep methods terms classification