Enhancing Early Brain Tumour Diagonsis Using Active Contour Algorithm Integrated With Swin Preprocessing And Canny Edge Detection

Abstract

Early and accurate detection of brain tumour is critical for effective treatment planning and improved patient prognosis. Traditional image segmentation techniques often struggle with low-resolution or noisy medical images, leading to imprecise localization of tumour boundaries. To address these limitations, this research proposes a novel framework that enhances the performance of classical contour-based segmentation by integrating SwinIR (Swin Transformer for Image Restoration) with Canny edge detection and Active Contour Models (Snakes). SwinIR, a transformer-based image enhancement technique, improves image clarity and resolution by capturing hierarchical contextual features, enabling better identification of tumour structures. The enhanced images are further refined using Canny edge detection and segmented via active contours to isolate tumour regions effectively.

The proposed method is validated using 10 brain CT images from the BraTS 2021 dataset, and its performance is evaluated using Pixel Accuracy and Abnormality Ratio. Experimental results demonstrate that SwinIR-preprocessed images yield the highest pixel accuracy of 97.12%, outperforming traditional approaches. Additionally, the framework improves the abnormality ratio by accurately highlighting tumour-affected regions. This integration of deep learning-based enhancement with classical segmentation offers a promising approach for more accurate and interpretable brain tumour diagnosis, paving the way for its application in clinical decision-support systems.