This study analyses intraoperative multispectral images taken from 47 brain tumour surgeries to investigate the diagnostic and surgical guidance potential of MSI. The research enrolled patients with various tumour types and introduces a hybrid model, uniting a transformer-coupled convolutional neural network (CNN), tailored for multispectral brain image segmentation. Leveraging MSI, the model was preliminarily assessed on ten meningioma and thirty-three glioma cases, each categorized into seven distinct classes. The model demonstrated a promising overall accuracies of 88.14% for meningioma and 85.64% for glioma. These initial results highlight the potential of the proposed hybrid architecture in multispectral brain image segmentation, laying the foundation for future research to optimize the model's performance with a larger patient cohort.
|