FM-Net: Focal Modulation-based Network forAccurate Skin Lesion Segmentation DOI Creative Commons
Asim Naveed, Syed S. Naqvi, Tariq M. Khan

et al.

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: March 28, 2025

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

Language: Английский

FM-Net: Focal Modulation-based Network forAccurate Skin Lesion Segmentation DOI Creative Commons
Asim Naveed, Syed S. Naqvi, Tariq M. Khan

et al.

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: March 28, 2025

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

Language: Английский

Citations

0