Skip to content
DSML.KZ
Publications

Book chapter · Routledge eBooks · November 17, 2025

A Double Attention-Aided U-Net Architecture for Skin Lesion and Histopathological Image Segmentation

Sujan Sarkar, Dipika Ray, Nikita Shiriaev, Dmitrii Kaplun, Ram Sarkar

Dmitry Kaplun
Open via DOI

Abstract

Detecting skin cancer from skin lesion images and segmenting nuclei in histopathological images are some of the critical tasks in medical image analysis. Early detection of skin cancer and accurate nucleus segmentation in histopathological images are crucial for diagnosing cancers such as breast, lung, and colorectal. However, skin lesion segmentation faces challenges, such as the variability in skin lesion images (size, shape, color, texture) and the presence of artifacts like hair and moles. Histopathological images are composed of cells that overlap, diverse textures, and varying shapes. Both tasks require substantial computational power for high-resolution segmentation. This chapter presents a novel U-Net model which utilizes MobileNetV2 as the encoder to extract key features from skin and histopathological images. The model incorporates wavelet textures (Haar) in the first encoder stage and uses Fused Attention blocks (combining Bit-Plane Attention and Efficient Channel Attention) in subsequent layers to enhance feature representation. The decoder upscales the segmentation map without direct connections between the encoder and decoder to preserve spatial details. Evaluated on ISIC 2016, MoNuSeg, and CPM-17 datasets, the model outperforms many existing models based on the dice score. The code of the proposed model is available at: https://www.w3.org/1999/xlink" xlink:href=" https://github.com/Cmatermedicalimageanalysis/DoubleAttentionUNet "> https://github.com/Cmatermedicalimageanalysis/DoubleAttentionUNet

AI in cancer detectionMedical Image Segmentation TechniquesCutaneous Melanoma Detection and Managementskin lesion segmentationhistopathological image segmentationnuclei segmentation

More by Dmitry Kaplun