Abstract:Breast cancer is a prevalent type of cancer that significantly impacts women health and stands as a leading cause of death. In clinical settings, automated breast tumor segmentation is crucial for early-stage diagnosis and treatment planning. Automatic tumor segmentation is complicated by several issues, including variations in tumor form, breast mass size, intensity variances, and a lack of training data. To address these issues, the proposed study emphasizes on multiscale feature extraction and transfer learning-based approach, utilizing past experiences from a source domain to improve the encoder design in the target domain. The multiscale feature fusion encoder yields precise location specific maps, successfully handling irregular shapes and sizes. Furthermore, the lightweight upsampling decoder enhances accurate restoration of tumor specific information. The inclusion of a sidewise deep supervision enhances breast lesion segmentation and facilitates easier localization in ultrasound images. Additionally, the utilization of a two-phase transfer learning and ensemble segmentation approach improves the quality of the segmentation map. Extensive experiments highlight the superiority of proposed model over state-of-the-art methods on three publicly available datasets. By utilizing publicly available databases, this work improves analytical perspectives on tumor types, methodologies and offers insightful information for the future research. The proposed system reached superior performance with dice score of 88.59% on BUSI and 92.11% on UDAIT dataset with outstanding tumor detection performance. Compared to previous deep learning models, this method reduces overfitting and greatly enhances generalization on smaller datasets. The identification of small tumor sizes, which is essential for early-stage diagnosis, is maximized by the proposed approach. This study enables precise segmentation of diverse tumor shapes and sizes, making it clinically useful for early-stage tumor diagnosis.

EfficientU-Net: A Novel Deep Learning Method for Breast Tumor Segmentation and Classification in Ultrasound Images

Breast Tumor Segmentation in Ultrasound Images Based on U-NET Model

Modifying the U-Net's Encoder-Decoder Architecture for Segmentation of Tumors in Breast Ultrasound Images

Rethinking the unpretentious U-net for medical ultrasound image segmentation

DAU-Net: Dual attention-aided U-Net for segmenting tumor in breast ultrasound images

U-Net and SegNet performances on lesion segmentation of breast ultrasonography images

Comparative Analysis of Segment Anything Model and U-Net for Breast Tumor Detection in Ultrasound and Mammography Images

ConnectedUNets++: Mass Segmentation from Whole Mammographic Images

Residual U-Network for Breast Tumor Segmentation from Magnetic Resonance Images

Inception UNet architecture for breast tumor segmentation and detection using hybrid deep learning approach

WATUNet: A Deep Neural Network for Segmentation of Volumetric Sweep Imaging Ultrasound

PDF-UNet: A semi-supervised method for segmentation of breast tumor images using a U-shaped pyramid-dilated network

Multimodal U-Net Breast Cancer Tumor Algorithm Based on Radio Frequency (RF) Data and Ultrasound Images

Mass segmentation in automated breast ultrasound using an enhanced attentive UNet

Advance Brain Tumor Segmentation Using Feature Fusion Methods with Deep U-Net Model with CNN for MRI Data

Tumor Segmentation in Breast Ultrasound Image by Means of Res Path Combined with Dense Connection Neural Network

UCapsNet: A Two-Stage Deep Learning Model Using U-Net and Capsule Network for Breast Cancer Segmentation and Classification in Ultrasound Imaging

Deep Hybrid Convolutional Neural Network for Segmentation of Melanoma Skin Lesion

Improving Breast Cancer Detection and Diagnosis through Semantic Segmentation Using the Unet3+ Deep Learning Framework

EfficientUNetViT: Efficient Breast Tumor Segmentation Utilizing UNet Architecture and Pretrained Vision Transformer

Attention based UNet model for breast cancer segmentation using BUSI dataset