Truly Generalizable Radiograph Segmentation With Conditional Domain Adaptation

Abstract

Digitization techniques for biomedical images yield disparate visual patterns in radiological exams. These pattern differences, which can be viewed as a domain-shift problem, may hamper the use of data-driven approaches for inference over these images, such as Deep Neural Networks. Another noticeable difficulty in this field is the lack of labeled data, even though in many cases there is an abundance of unlabeled data available. Therefore, an important step in improving the generalization capabilities of these methods and mitigate domain-shift effects is to perform unsupervised or semi-supervised adaptation between different domains of biomedical images. In this work, we propose a novel approach for segmentation of biomedical images based on Generative Adversarial Networks. The proposed method, named Conditional Domain Adaptation Generative Adversarial Network (CoDAGAN), merges unsupervised networks with supervised deep semantic segmentation architectures in order to create a semi-supervised method capable of learning from both unlabeled and labeled data, whenever labeling is available. We conducted experiments to compare our method with traditional and state-of-the-art baselines by using several domains, datasets, and segmentation tasks. The proposed method yielded consistently better results than the baselines in scarce labeled data scenarios, achieving Jaccard values greater than 0.9 and good segmentation quality in most tasks. Unsupervised Domain Adaptation results were observed to be close to the Fully Supervised Domain Adaptation used in the traditional procedure of fine-tuning pretrained networks.

OdontoAI: A human-in-the-loop labeled data set and an online platform to boost research on dental panoramic radiographs

“…They employed a Mask R-CNN (He et al, 2017) trained over binary masks that separated the teeth from the background and showed that their approach outperformed traditional solutions to the task. The authors also made their data public under the name UFBA-UESC Dental Image Data Set 3 , which proved to be a valuable resource as it has been extensively used by many works (Koch et al, 2019;Zhao et al, 2020;Oliveira et al;Chen et al, 2021;Cui et al, 2021;Hsu and Wang, 2021). In an extension of Silva et al's work, Jader et al (2018) segmented tooth instances on radiographs of the UFBA-UESC Dental Image Data Set also using Mask R-CNN, though not numbering them.…”

Domain Adaptation Deep Attention Network for Automatic Logo Detection and Recognition in Google Street View

“…Moreover, domain adaptation could adapt domain shifts in appearance and semantic levels of an object to achieve higher accuracy and effectiveness [24][25]. Hugo et al [26] proposed a new method of domain adaptation named the Conditional Domain Adaptation Generative Adversarial Network (CoDAGAN) for segmentation purposes in biomedical images. It merges unsupervised [27] and supervised networks to become a semisupervised method that can learn from unlabeled and labeled data.…”

Dynamic boundary‐insensitive loss for magnetic resonance medical image segmentation

“…Cycle‐GAN 37 and its variants are frequently integrated into domain adaptation methods to address the domain shift. Oliveira et al 38 . proposed the adversarial‐networks based domain adaptation method to learn from both labeled and unlabeled data.…”