Haeyun Lee

Accurate trajectory estimation using low-cost inertial measurement unit (IMU) data remains a challenging task due to sensor noise, bias, and drift. In this paper, we propose the Convolutional IMU Transformer (CoIT), a deep neural network that combines Convolutional Token Embedding with a hierarchical Transformer architecture to effectively capture both local motion features and global temporal dependencies from raw IMU measurements. The proposed network is designed to overcome the limitations of traditional dead reckoning methods and prior learning-based approaches that rely heavily on inaccurate orientation estimates. CoIT further incorporates a multi-stage Convolutional Projection and downsampling strategy to improve computational efficiency while preserving representational capacity. We evaluate CoIT on three large-scale public datasets (RIDI, RoNIN, and NeurIT) and demonstrate consistent performance gains over existing state-of-the-art models. On the RoNIN dataset, CoIT achieves an absolute trajectory error (ATE) of 3.39 ± 0.59 m, a time-normalized relative trajectory error (T-RTE) of 2.63 ± 0.57 m, and a distance-normalized relative trajectory error (D-RTE) of 0.17 ± 0.02 m, while also demonstrating robust performance on unseen subjects and competitive cross-domain transfer from pedestrian motion (RoNIN) to indoor robot motion (NeurIT) without fine-tuning. Furthermore, an ablation study on model complexity reveals a clear trade-off between accuracy and latency, demonstrating the architecture’s scalability for practical deployment. These results highlight the robustness and statistical reliability of the proposed model within the evaluated public benchmarks in GPS-denied settings.

Medical ultrasound imaging is a key diagnostic tool across various fields, with computer-aided diagnosis systems benefiting from advances in deep learning. However, its lower resolution and artifacts pose challenges, particularly for non-specialists. The simultaneous acquisition of degraded and high-quality images is infeasible, limiting supervised learning approaches. Additionally, self-supervised and zero-shot methods require extensive processing time, conflicting with the real-time demands of ultrasound imaging. Therefore, to address the aforementioned issues, we propose real-time ultrasound image enhancement via a self-supervised learning technique and a test-time adaptation for sophisticated rotational cuff tear diagnosis. The proposed approach learns from other domain image datasets and performs self-supervised learning on an ultrasound image during inference for enhancement. Our approach not only demonstrated superior ultrasound image enhancement performance compared to other state-of-the-art methods but also achieved an 18% improvement in the RCT segmentation performance.

For several decades, the significance of geospatial object segmentation in remote-sensing images has been emphasized for both scientific and industrial purposes. Particularly, the importance of granularity in the boundaries of geospatial objects, called fine-grained segmentation, has been highlighted in achieving the precise identification and classification of objects. Deep learning-based segmentation methodologies, surpassing the limitations of the conventional vision-based analysis, have yielded accurate predictions by utilizing Convolutional Neural Networks (CNNs). However, CNNs classify images pixel-wise and produce outputs based on probability distributions derived from the SoftMax function. This approach precludes the reflection of morphological properties, such as shape and object density, during predictions in remote sensing imagery, leading to imprecise results. Additionally, due to the intrinsic attributes of probability-based segmentation, fine-grained segmentation may not be achieved, leading to coarse predictions in the boundaries of geospatial objects. To address this issue, this paper introduces a novel deep learning framework, the Density-based Guide Network (DG-Net), which incorporates the density of segmentation targets into pixel-wise classification through a test-time adaptation learning methodology. DG-Net first recognizes the density of segmentation targets in the input images, then fine-tunes the baseline network to reflect this density, generating precise segmentation outputs. The efficacy of DG-Net is demonstrated through various multi-target segmentation benchmarks in remote sensing imagery, with experimental results demonstrating superior performance compared to state-of-the-art models in object segmentation across numerous aerial image and satellite image datasets.

The simulation of human neurons and neurotransmission mechanisms has been realized in deep neural networks based on the theoretical implementations of activation functions. However, recent studies have reported that the threshold potential of neurons exhibits different values according to the locations and types of individual neurons, and that the activation functions have limitations in terms of representing this variability. Therefore, this study proposes a simple yet effective activation function that facilitates different thresholds and adaptive activations according to the positions of units and the contexts of inputs. Furthermore, the proposed activation function mathematically exhibits a more generalized form of Swish activation function, and thus we denoted it as Adaptive SwisH (ASH). ASH highlights informative features that exhibit large values in the top percentiles in an input, whereas it rectifies low values. Most importantly, ASH exhibits trainable, adaptive, and context-aware properties compared to other activation functions. Furthermore, ASH represents general formula of the previously studied activation function and provides a reasonable mathematical background for the superior performance. To validate the effectiveness and robustness of ASH, we implemented ASH into many deep learning models for various tasks, including classification, detection, segmentation, and image generation. Experimental analysis demonstrates that our activation function can provide the benefits of more accurate prediction and earlier convergence in many deep learning applications.

The performance of computer-aided diagnosis (CAD) systems that are based on ultrasound imaging has been enhanced owing to the advancement in deep learning. However, because of the inherent speckle noise in ultrasound images, the ambiguous boundaries of lesions deteriorate and are difficult to distinguish, resulting in the performance degradation of CAD. Although several methods have been proposed to reduce speckle noise over decades, this task remains a challenge that must be improved to enhance the performance of CAD. In this article, we propose a deep content-aware image prior (DCAIP) with a content-aware attention module (CAAM) for superior despeckling of ultrasound images without clean images. For the image prior, we developed a CAAM to deal with the content information in an input image. In this module, super-pixel pooling (SPP) is used to give attention to salient regions in an ultrasound image. Therefore, it can provide more content information regarding the input image when compared to other attention modules. The DCAIP consists of deep learning networks based on this attention module. The DCAIP is validated by applying it as a preprocessing step for breast tumor segmentation in ultrasound images, which is one of the tasks in CAD. Our method improved the segmentation performance by 15.89% in terms of the area under the precision–recall (PR) curve (AUPRC). The results demonstrate that our method enhances the quality of ultrasound images by effectively reducing speckle noise while preserving important information in the image, promising for the design of superior CAD systems.

Various deep learning-based segmentation models have been developed to segment buildings in aerial images. However, the segmentation maps predicted by the conventional convolutional neural network-based methods cannot accurately determine the shapes and boundaries of segmented buildings. In this article, to improve the prediction accuracy for the boundaries and shapes of segmented buildings in aerial images, we propose the boundary-oriented binary building segmentation model (B3SM). To construct the B3SM for boundary-enhanced semantic segmentation, we present two-scheme learning (Schemes I and II), which uses the upsampling interpolation method (USIM) as a new operator and a boundary-oriented loss function (B-Loss). In Scheme I, a raw input image is processed and transformed into a presegmented map. In Scheme II, the presegmented map from Scheme I is transformed into a more fine-grained representation. To connect these two schemes, we use the USIM operator. In addition, the novel B-Loss function is implemented in B3SM to extract the features of the boundaries of buildings effectively. To perform quantitative evaluation of the shapes and boundaries of segmented buildings generated by B3SM, we develop a new metric called the boundary-oriented intersection over union (B-IoU). After evaluating the effectiveness of two-scheme learning, USIM, and B-Loss for building segmentation, we compare the performance of B3SM to those of other state-of-the-art methods using public and custom datasets. The experimental results demonstrate that the B3SM outperforms other state-of-the-art models, resulting in more accurate shapes and boundaries for segmented buildings in aerial images.

Background: Federated learning is a decentralized approach to machine learning; it is a training strategy that overcomes medical data privacy regulations and generalizes deep learning algorithms. Federated learning mitigates many systemic privacy risks by sharing only the model and parameters for training, without the need to export existing medical data sets. In this study, we performed ultrasound image analysis using federated learning to predict whether thyroid nodules were benign or malignant. Objective: The goal of this study was to evaluate whether the performance of federated learning was comparable with that of conventional deep learning. Methods: A total of 8457 (5375 malignant, 3082 benign) ultrasound images were collected from 6 institutions and used for federated learning and conventional deep learning. Five deep learning networks (VGG19, ResNet50, ResNext50, SE-ResNet50, and SE-ResNext50) were used. Using stratified random sampling, we selected 20% (1075 malignant, 616 benign) of the total images for internal validation. For external validation, we used 100 ultrasound images (50 malignant, 50 benign) from another institution. Results: For internal validation, the area under the receiver operating characteristic (AUROC) curve for federated learning was between 78.88% and 87.56%, and the AUROC for conventional deep learning was between 82.61% and 91.57%. For external validation, the AUROC for federated learning was between 75.20% and 86.72%, and the AUROC curve for conventional deep learning was between 73.04% and 91.04%. Conclusions: We demonstrated that the performance of federated learning using decentralized data was comparable to that of conventional deep learning using pooled data. Federated learning might be potentially useful for analyzing medical images while protecting patients’ personal information.

Change detection is an important task in the field of remote sensing. Various change detection methods based on convolutional neural networks (CNNs) have recently been proposed for remote sensing using satellite or aerial images. However, existing methods allow only the partial use of content information in images during change detection because they adopt simple feature similarity measurements or pixel-level loss functions to construct their network architectures. Therefore, when these methods are applied to complex urban areas, their performance in terms of change detection tends to be limited. In this article, a novel CNN-based change detection approach, referred to as a local similarity Siamese network (LSS-Net), with a cosine similarity measurement, was proposed for better urban land change detection in remote sensing images. To use content information on two sequential images, a new change attention map-based content loss function was developed in this study. In addition, to enhance the performance of the LSS-Net in terms of change detection, a suitable feature similarity measurement method, incorporated into a local similarity attention module, was determined through systemic experiments. To verify the change detection performance of the LSS-Net, it was compared with other state-of-the-art methods. The experimental results show that the proposed method outperforms the state-of-the-art methods in terms of the F1 score (0.9630, 0.9377, and 0.7751) and kappa (0.9581, 0.9351, and 0.7646) on the three test datasets, thus suggesting its potential for various remote sensing applications.

BACKGROUND Incomplete Kawasaki disease (KD) has often been misdiagnosed due to a lack of the clinical manifestations of classic KD. However, it is associated with a markedly higher prevalence of coronary artery lesions. Identifying coronary artery lesions by echocardiography is important for the timely diagnosis of and favorable outcomes in KD. Moreover, similar to KD, coronavirus disease 2019, currently causing a worldwide pandemic, also manifests with fever; therefore, it is crucial at this moment that KD should be distinguished clearly among the febrile diseases in children. OBJECTIVE In this study, we aimed to validate a deep learning algorithm for classification of KD and other acute febrile diseases. METHODS We obtained coronary artery images by echocardiography of children (n = 88 for KD; n = 65 for pneumonia). We trained six deep learning networks (VGG19, Xception, ResNet50, ResNext50, SE-ResNet50, and SE-ResNext50) using the collected data. RESULTS SE-ResNext50 showed the best performance in terms of accuracy, specificity, and precision in the classification. SE-ResNext50 offered a sensitivity of 82.64% and a specificity of 58.12%. CONCLUSIONS The results of our study suggested that deep learning algorithms have similar performance to an experienced cardiologist in detecting coronary artery lesions to facilitate the diagnosis of KD. CLINICALTRIAL none