FM-Net: Focal Modulation-based Network forAccurate Skin Lesion Segmentation

Precise segmentation of skin lesions in dermoscopic images is critical for skin cancers, including melanoma, as accurate delineation is essential for timely and effective diagnosis. Cancerous skin lesions, particularly malignant melanoma, significantly contribute to the mortality rate underscoring the need for early detection and precise diagnostic techniques. However, achieving this precision poses challenges due to indistinct lesion borders, asymmetrical shapes, and common obstructions like hair, markings, and occlusions. This study addresses these challenges by introducing an end-to-end trainable network incorporating focal modulation to enhance feature refinement for pixel-level classification. The focal modulation captures fine-grained multi-scale features and contextual information for precise segmentation of lesions. The decoder part of the proposed method utilizes transposed convolution for up-sampling, which preserves the spatial detail necessary for high-resolution segmentation. The proposed method achieves state-of-the-art (SOTA) performance in skin and breast lesion segmentation, validated across multiple benchmark datasets. An outstanding feature of the proposed method is its ability to deliver performance without employing data augmentation. The robustness of the proposed method is demonstrated by its performance on the ISIC datasets, achieving Jaccard index scores of 89.60% on ISIC 2016, 82.34% on ISIC 2017, and 87.71% on ISIC 2018. Moreover, the performance of the proposed method is computed on the breast ultrasound lesion images (BUSI) dataset. The comprehensive performance of our method highlights its ability to accurately segment lesions and its potential to assist in timely diagnosis. The code to reproduce the results is made available at \href{https://github.com/Asim-Naveed/FM-Nets}{GitHub}.