TrEnD: A transformer‐based encoder‐decoder model with adaptive patch embedding for mass segmentation in mammograms

Abstract

Background: Breast cancer is one of the most prevalent malignancies diagnosed in women. Mammogram inspection in the search and delineation of breast tumors is an essential prerequisite for a reliable diagnosis. However, analyzing mammograms by radiologists is time-consuming and prone to errors. Therefore, the development of computer-aided diagnostic (CAD) systems to automate the mass segmentation procedure is greatly expected. Purpose: Accurate breast mass segmentation in mammograms remains challenging in CAD systems due to the low contrast, various shapes, and fuzzy boundaries of masses. In this paper, we propose a fully automatic and effective mass segmentation model based on deep learning for improving segmentation performance. Methods:We propose an effective transformer-based encoder-decoder model (TrEnD). Firstly, we introduce a lightweight method for adaptive patch embedding (APE) of the transformer, which utilizes superpixels to adaptively adjust the size and position of each patch. Secondly, we introduce a hierarchical transformer-encoder and attention-gated-decoder structure, which is beneficial for progressively suppressing interference feature activations in irrelevant background areas. Thirdly, a dual-branch design is employed to extract and fuse globally coarse and locally fine features in parallel, which could capture the global contextual information and ensure the relevance and integrity of local information. The model is evaluated on two public datasets CBIS-DDSM and INbreast. To further demonstrate the robustness of TrEnD, different cropping strategies are applied to these datasets, termed tight, loose, maximal, and mixframe.Finally,ablation analysis is performed to assess the individual contribution of each module to the model performance. Results:The proposed segmentation model provides a high Dice coefficient and Intersection over Union (IoU) of 92.20% and 85.81% on the mix-frame CBIS-DDSM, while 91.83% and 85.29% for the mix-frame INbreast, respectively. The segmentation performance outperforms the current state-of -the-art approaches. By adding the APE and attention-gated module, the Dice and IoU have improved by 6.54% and 10.07%. Conclusion: According to extensive qualitative and quantitative assessments, the proposed network is effective for automatic breast mass segmentation, and has adequate potential to offer technical assistance for subsequent clinical diagnoses.

A Systematic Literature Review on Mammography: Deep Learning in Redefining Breast Cancer Diagnosis for the Asian Perspective

“…Adaptive Neuro Fuzzy Classifier (Rani et al, 2023) Due to its structure, which incorporates the intuitive nature of fuzzy logic systems with the learning capabilities of neural networks, this approach is employed for handling complex relationships in data. (Alkhaleefah et al, 2022;Abdelhafiz et al, 2020;Ashwini et al, 2023;Bai et al, 2022;Boudouh and Bouakkaz, 2023a;Lingampally and Kavuri, 2023;Liu et al, 2023;Mahmood et al, 2021Mahmood et al, , 2022Malathi and Latha, 2023;Maqsood et al, 2022;Cao et al, 2020;Castro-Tapia et al, 2021;Ertugrul and Abdullah, 2022;El Houby and Yassin, 2021;Falconi et al, 2020;George Melekoodappattu et al, 2022;Gerbasi et al, 2023;Gnanasekaran et al, 2020;Harris et al, 2023;Montaha et al, 2021;Huang and Lin, 2021;Jayandhi et al, 2022;Karthiga et al, 2022 (Cai et al, 2021;Narayanan et al, 2022;Nithya and Santhi, 2021;Oyelade and Ezugwu, 2022;Oza et al, 2022;Pawar et al, 2022;Ragab et al, 2021Ragab et al, , 2019Ravikumar et al, 2023;Kavitha et al, 2021) Histogram Equalization (Ahmad et al, 2023;Alfifi et al, 2020;Arora et al, 2020;Babu and Jerome, 2022;Baccouche et al, 2022;…”

“…Adding a super-pixel average pooling layer to the cGAN decoders improves boundary segmentation. Transformer based Encoder Decoder model (Liu et al, 2023) The transformer-encoder represents the input, whereas an attention-gated-decoder framework enhances the outcome, enabling for a hierarchical and enhanced approach to capturing complex relationships and patterns in the data. ESP-Net (Shanker and Vadivel, 2022) ESPNet breaks down the traditional convolutional operation into point-wise convolutions and a spatial pyramid of dilated convolutions to reduce processing time.…”