Prototype-Based Approach for One-Shot Segmentation of Brain Tumors using Few-Shot Learning

The potential for augmenting the segmentation of brain tumors through the useof few-shot learning is vast. Although several deep learning networks (DNNs)demonstrate promising results in terms of segmentation, they require asubstantial quantity of training data in order to produce suitable outcomes.Furthermore, a major issue faced by most of these models is their ability toperform well when faced with unseen classes. To address these challenges, wepropose a one-shot learning model for segmenting brain tumors in magneticresonance images (MRI) of the brain, based on a single prototype similarityscore. Leveraging the recently developed techniques of few-shot learning, whichinvolve the utilization of support and query sets of images for training andtesting purposes, we strive to obtain a definitive tumor region by focusing onslices that contain foreground classes. This approach differs from other recentDNNs that utilize the entire set of images. The training process for this modelis carried out iteratively, with each iteration involving the selection ofrandom slices that contain foreground classes from randomly sampled data as thequery set, along with a different random slice from the same sample as thesupport set. In order to distinguish the query images from the classprototypes, we employ a metric learning-based approach that relies onnon-parametric thresholds. We employ the multimodal Brain Tumor ImageSegmentation (BraTS) 2021 dataset, which comprises 60 training images and 350testing images. The effectiveness of the model is assessed using the mean dicescore and mean Intersection over Union (IoU) score.