SF2Former: Amyotrophic lateral sclerosis identification from multi-center MRI data using spatial and frequency fusion transformer

• DOI: 10.1016/j.compmedimag.2023.102279
• Publication Date: 2023-07-29T23:33:08Z
• Authors: Rafsanjany Kushol; Collin C. Luk; Avyarthana Dey; Michael Benatar; Hannah Briemberg; Annie Dionne; Nicolas Dupré; Richard Frayne; Angela Genge; Summer Gibson; Simon J. Graham; Lawrence Korngut; Peter Seres; Robert C. Welsh; Alan H. Wilman; Lorne Zinman; Sanjay Kalra; Yee-Hong Yang
• Keywords: Discriminative model

Amyotrophic Lateral Sclerosis (ALS) is a complex neurodegenerative disorder characterized by motor neuron degeneration. Significant research has begun to establish brain magnetic resonance imaging (MRI) as potential biomarker diagnose and monitor the state of disease. Deep learning emerged prominent class machine algorithms in computer vision shown successful applications various medical image analysis tasks. However, deep methods applied neuroimaging have not achieved superior performance classifying ALS patients from healthy controls due insignificant structural changes correlated with pathological features. Thus, critical challenge models identify discriminative features limited training data. To address this challenge, study introduces framework called SF2Former, which leverages power transformer architecture distinguish subjects control group exploiting long-range relationships among Additionally, spatial frequency domain information combined enhance network's performance, MRI scans are initially captured then converted domain. The proposed trained using series consecutive coronal slices utilizes pre-trained weights ImageNet through transfer learning. Finally, majority voting scheme employed on each subject generate final classification decision. extensively evaluated multi-modal data (i.e., T1-weighted, R2*, FLAIR) two well-organized versions Canadian Neuroimaging Consortium (CALSNIC) multi-center datasets. experimental results demonstrate superiority strategy terms accuracy compared several popular learning-based techniques.