RCA-IUnet: A residual cross-spatial attention guided inception U-Net model for tumor segmentation in breast ultrasound imaging

TL;DR

This paper introduces RCA-IUnet, a novel deep learning model with residual inception and cross-spatial attention for improved breast tumor segmentation in ultrasound images, achieving superior accuracy with minimal training parameters.

Contribution

The paper presents a new residual cross-spatial attention guided inception U-Net model that enhances tumor segmentation performance in breast ultrasound images, outperforming existing models.

Findings

Outperformed state-of-the-art segmentation models on public datasets.

Achieved high accuracy with minimal training parameters.

Effectively segmented tumors of varying sizes.

Abstract

The advancements in deep learning technologies have produced immense contributions to biomedical image analysis applications. With breast cancer being the common deadliest disease among women, early detection is the key means to improve survivability. Medical imaging like ultrasound presents an excellent visual representation of the functioning of the organs; however, for any radiologist analysing such scans is challenging and time consuming which delays the diagnosis process. Although various deep learning based approaches are proposed that achieved promising results, the present article introduces an efficient residual cross-spatial attention guided inception U-Net (RCA-IUnet) model with minimal training parameters for tumor segmentation using breast ultrasound imaging to further improve the segmentation performance of varying tumor sizes. The RCA-IUnet model follows U-Net topology with residual inception depth-wise separable convolution and hybrid pooling (max pooling and spectral pooling) layers. In addition, cross-spatial attention filters are added to suppress the irrelevant features and focus on the target structure. The segmentation performance of the proposed model is validated on two publicly available datasets using standard segmentation evaluation metrics, where it outperformed the other state-of-the-art segmentation models.