Purpose: To propose a neural network (NN) that can effectively segment orbital tissue in computed tomography (CT) images of Graves’ orbitopathy (GO) patients. Methods: We analyzed orbital CT scans from 701 GO patients diagnosed between 2010 and 2019 and devised an effective NN specializing in semantic orbital tissue segmentation in GO patients’ CT images. After four conventional (Attention U-Net, DeepLab V3+, SegNet, and HarDNet-MSEG) and the proposed NN train the various manual orbital tissue segmentations, we calculated the Dice coefficient and Intersection over Union for comparison. Results: CT images of the eyeball, four rectus muscles, the optic nerve, and the lacrimal gland tissues from all 701 patients were analyzed in this study. In the axial image with the largest eyeball area, the proposed NN achieved the best performance, with Dice coefficients of 98.2% for the eyeball, 94.1% for the optic nerve, 93.0% for the medial rectus muscle, and 91.1% for the lateral rectus muscle. The proposed NN also gave the best performance for the coronal image. Our qualitative analysis demonstrated that the proposed NN outputs provided more sophisticated orbital tissue segmentations for GO patients than the conventional NNs. Conclusion: We concluded that our proposed NN exhibited an improved CT image segmentation for GO patients over conventional NNs designed for semantic segmentation tasks.

Thyroid-associated orbitopathy is an autoimmune disease that causes changes in various structures close to the eye. Medical images, such as three-dimensional computed tomography scans, can be used by medical experts to diagnose thyroid-associated orbitopathy. Meanwhile, image segmentation has been widely used in medical imaging owing to its significant impact on improving model performance by filtering out unnecessary pixel values. In this study, a neural network specialized in processing multiple segmented images was proposed to evaluate thyroid orbitopathy activity, focusing on the fact that multiple segmented images can be extracted from orbital computed tomography scans. The proposed neural network consists of multiple convolutional embedding heads, a group squeeze-and-excitation block, and a classifier stage. Our empirical study shows that the proposed model outperforms four baseline models on a thyroid-associated orbitopathy activity dataset obtained from a cohort of 1,068 patients at Chung-Ang University Hospital between January 2008 and October 2019. The proposed model achieved an average area under the receiver operating characteristic curve of 0.800, accuracy of 0.721, F1 score of 0.416, sensitivity of 0.728, and specificity of 0.720 across 50 replicate experiments. The source code for the proposed model is available at https://github.com/tkdgur658/MultiheadGroupSENet.

Graves’ orbitopathy is an inflammatory disorder that causes changes in different structures close to the eye. Accurate and consistent diagnoses are essential to improve the quality of life for Graves’ orbitopathy patients. To this end, a number of studies on Graves’ orbitopathy have been conducted based on neural networks recently. However, treatment decision methods based on neural networks have been much less addressed. This study aims to propose an effective deep neural network-based diagnosis method that makes treatment decisions for Graves’ orbitopathy patients. Specifically, the proposed method adopts a high-resolution feature enhancement and low-resolution feature preservation strategy focusing on the following points. First, the loss of detailed spatial information during the alignment of pixel spacing in computed tomography images leads to a decrease in performance. Thus, we preserve the detailed information of the images through high-resolution resampling. Second, existing studies lack sophistication in network design. The baseline network was improved by four modifications. Finally, resizing and coarse cropping cause learning instability. Thus, we adopt padding and fine-grained cropping. Our empirical study shows that the proposed method outperforms two existing neural network-based Graves’ orbitopathy diagnostic pipelines achieving an average area under the receiver operating characteristic curve of 0.793, accuracy of 0.699, F1 score of 0.416, sensitivity of 0.723, and specificity of 0.694 in five repetitive experiments. Furthermore, in-depth analysis provides several future research directions in computed tomography preprocessing and deep neural network design. The source code for the proposed model is available at https://github.com/tkdgur658/GOTDNet .

Crack segmentation plays a vital role in ensuring structural safety, yet practical deployment on resource-limited platforms demands models that balance accuracy with efficiency. While high-accuracy models often rely on computationally heavy designs to expand their receptive fields, recent lightweight approaches typically delay this expansion to the deepest, low-resolution layers to maintain efficiency. This design choice leaves long-range context underutilized, where fine-grained evidence is most intact. In this paper, we propose an evidence-preserving receptive-field expansion network, which integrates a multi-scale dilated block to efficiently capture long-range context from the earliest stages and an input-guided gate that leverages grayscale conversion, average pooling, and gradient extraction to highlight crack evidence directly from raw inputs. Experiments on six benchmark datasets demonstrate that the proposed network achieves consistently higher accuracy under lightweight constraints. Each of the three proposed variants—Base, Small, and Tiny—outperforms its corresponding baselines with larger parameter counts, surpassing a total of 13 models. For example, the Base variant reduces parameters by 66% compared to the second-best CrackFormer II and floating-point operations by 53% on the Ceramic dataset, while still delivering superior accuracy. Pareto analyses further confirm that the proposed model establishes a superior accuracy–efficiency trade-off across parameters and floating-point operations.

Accurate segmentation of the posterior extraocular muscles is important for diagnosing thyroid-associated orbitopathy. Given that this region occupies a small area in computed tomography images, effectively leveraging low-level information within high-resolution feature maps is crucial. However, such information is often lost in conventional networks because of the excessive expansion of the receptive field in multi-scale blocks, leading to less accurate pixel-wise classification. To address these challenges, we propose a layer-wise residual-learning approach for multi-scale blocks. The residual connections are integrated between each layer within the stacked convolutional layers of the multi-scale block, thereby preserving low-level information and enhancing the high-level feature extraction for small objects. The proposed model was evaluated using computed tomography images from 232 patients at Chung-Ang University Hospital (2009–2022) and compared with 11 state-of-the-art networks. Four datasets corresponding to the medial, lateral, superior, and inferior rectus muscles of the left eye, using six measures, were used for this evaluation. The results showed that proposed model achieved the highest performance in 15 of the 24 cases and the second-highest performance in eight cases. Our ablation study and qualitative analysis demonstrate the efficacy of the proposed model in segmenting small muscle regions.