Convolutional layer exertion on few-shot learning for brain tumor classification

Brain tumors, though relatively rare, pose a significant threat due to their critical location within the brain, impacting essential bodily functions. Accurate and timely diagnosis is vital, but traditional diagnostic methods are time-intensive and rely heavily on large labeled datasets. This study addresses these challenges by proposing a Few-Shot Learning (FSL) framework enhanced with Convolutional Neural Networks (CNNs) to classify brain tumors using MRI images. By employing the Matching Network architecture, the model leverages limited training data through an N-way-K-shot setup. Training results demonstrated accuracy levels of 71.58% (1-shot) and 82.89% (5-shot) for 1-layer CNNs, 66.65% (1-shot) and 84.03% (5-shot) for 3-layer CNNs, and 63.43% (1-shot) and 84.94% (5-shot) for 5-layer CNNs. However, validation accuracy revealed overfitting concerns, with the highest performance at 51.56% (1-layer, 1-shot). These results underscore the potential of FSL in medical imaging while highlighting the need for advanced augmentation and feature representation techniques to improve generalization.