The purpose of this study was to directly and quantitatively measure BMD from Cone-beam CT (CBCT) images by enhancing the linearity and uniformity of the bone intensities based on a hybrid deep-learning model (QCBCT-NET) of combining the generative adversarial network (Cycle-GAN) and U-Net, and to compare the bone images enhanced by the QCBCT-NET with those by Cycle-GAN and U-Net. We used two phantoms of human skulls encased in acrylic, one for the training and validation datasets, and the other for the test dataset. We proposed the QCBCT-NET consisting of Cycle-GAN with residual blocks and a multi-channel U-Net using paired training data of quantitative CT (QCT) and CBCT images. The BMD images produced by QCBCT-NET significantly outperformed the images produced by the Cycle-GAN or the U-Net in mean absolute difference (MAD), peak signal to noise ratio (PSNR), normalized cross-correlation (NCC), structural similarity (SSIM), and linearity when compared to the original QCT image. The QCBCT-NET improved the contrast of the bone images by reflecting the original BMD distribution of the QCT image locally using the Cycle-GAN, and also spatial uniformity of the bone images by globally suppressing image artifacts and noise using the two-channel U-Net. The QCBCT-NET substantially enhanced the linearity, uniformity, and contrast as well as the anatomical and quantitative accuracy of the bone images, and demonstrated more accuracy than the Cycle-GAN and the U-Net for quantitatively measuring BMD in CBCT.

We developed an automatic method for staging periodontitis on dental panoramic radiographs using the deep learning hybrid method. A novel hybrid framework was proposed to automatically detect and classify the periodontal bone loss of each individual tooth. The framework is a hybrid of deep learning architecture for detection and conventional CAD processing for classification. Deep learning was used to detect the radiographic bone level (or the CEJ level) as a simple structure for the whole jaw on panoramic radiographs. Next, the percentage rate analysis of the radiographic bone loss combined the tooth long-axis with the periodontal bone and CEJ levels. Using the percentage rate, we could automatically classify the periodontal bone loss. This classification was used for periodontitis staging according to the new criteria proposed at the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. The Pearson correlation coefficient of the automatic method with the diagnoses by radiologists was 0.73 overall for the whole jaw (p < 0.01), and the intraclass correlation value 0.91 overall for the whole jaw (p < 0.01). The novel hybrid framework that combined deep learning architecture and the conventional CAD approach demonstrated high accuracy and excellent reliability in the automatic diagnosis of periodontal bone loss and staging of periodontitis.

The blood glucose prediction algorithms were widely studied for the patients with type 1 diabetes. We analyzed the data of 16 type 2 diabetic patients underwent continuous glucose monitoring (CGM) through deep-learning models. Feed Forward Neural Network (FFNN) and Long Short-Term Memory (LSTM) were used to model and predict the blood glucose. Blood glucose levels after 5, 15, 30, and 45 min were predicted for each patient. To evaluate the predictive power of model, the Root Mean Square Error (RMSE) was calculated by comparing the actual blood glucose and predictive blood glucose. The value of RMSE by FFNN was lower than that by LSTM. The predictive success rates for the glycemic excursion at 5 minutes by FFNN, LSTM were 96% and 95%, the predictive power of two models gradually decreased with time (Figure 1). These results suggest that the well trained deep-learning model would be possible to predict the harmful glycemic events among type 2 diabetic patient. Disclosure D. Kim: None. H. Lee: None. Y. Kim: None. S. Kim: None. S. Lee: None. S. Chun: None.

Automated Optical Inspection (AOI) of Printed Circuit Boards (PCBs) suffers from scarce labeled data and frequent domain shifts caused by variations in camera optics, illumination, and product design. These limitations hinder the development of accurate and reliable deep-learning models in manufacturing settings. To address this challenge, this study systematically benchmarks three Parameter-Efficient Fine-Tuning (PEFT) strategies-Linear Probe, Low-Rank Adaptation (LoRA), and Visual Prompt Tuning (VPT)-applied to two representative foundation vision models: the Contrastive Language-Image Pretraining Vision Transformer (CLIP-ViT-B/16) and the Self-Distillation with No Labels Vision Transformer (DINOv2-S/14). The models are evaluated on six-class PCB defect classification tasks under few-shot (k = 5, 10, 20) and full-data regimes, analyzing both performance and reliability. Experiments show that VPT achieves 0.99 ± 0.01 accuracy and 0.998 ± 0.001 macro-Area Under the Precision-Recall Curve (macro-AUPRC), reducing classification error by approximately 65% compared with Linear and LoRA while tuning fewer than 1.5% of backbone parameters. Reliability, assessed by the stability of precision-recall behavior across different decision thresholds, improved as the number of labeled samples increased. Furthermore, class-wise and few-shot analyses revealed that VPT adapts more effectively to rare defect types such as Spur and Spurious Copper while maintaining near-ceiling performance on simpler categories (Short, Pinhole). These findings collectively demonstrate that prompt-based adaptation offers a quantitatively favorable trade-off between accuracy, efficiency, and reliability. Practically, this positions VPT as a scalable strategy for factory-level AOI, enabling the rapid deployment of robust defect inspection models even when labeled data is scarce.

Background: Severe class imbalance limits reliable tabular AI in manufacturing, finance, and healthcare. Methods: We built a modular pipeline comprising correlation-aware seriation; a hybrid convolutional neural network (CNN)–transformer–Bidirectional Long Short-Term Memory (BiLSTM) encoder; variational autoencoder (VAE)-based minority augmentation; and deep/tree ensemble heads (XGBoost and Support Vector Machine, SVM). We benchmarked the Synthetic Minority Oversampling Technique (SMOTE) and ADASYN under identical protocols. Focal loss and ensemble weights were tuned per dataset. The primary metric was the Area Under the Precision–Recall Curve (AUPRC), with receiver operating characteristic area under the curve (ROC AUC) as complementary. Synthetic-data fidelity was quantified by train-on-synthetic/test-on-real (TSTR) utility, two-sample discriminability (ROC AUC of a real-vs-synthetic classifier), and Maximum Mean Discrepancy (MMD2). Results: Across five datasets (SECOM, CREDIT, THYROID, APS, and UCI), augmentation was data-dependent: VAE led on APS (+3.66 pp AUPRC vs. SMOTE) and was competitive on CREDIT (+0.10 pp vs. None); the SMOTE dominated SECOM; no augmentation performed best for THYROID and UCI. Positional embedding (PE) with seriation helped when strong local correlations were present. Ensembles typically favored XGBoost while benefiting from the hybrid encoder. Efficiency profiling and a slim variant supported latency-sensitive use. Conclusions: A data-aware recipe emerged: prefer VAE when fidelity is high, the SMOTE on smoother minority manifolds, and no augmentation when baselines suffice; apply PE/seriation selectively and tune per dataset for robust, reproducible deployment.

(1) Background. Printed circuit board (PCB) inspection is increasingly constrained by the cost and latency of reliable labels, owing to tiny/low-contrast defects embedded in complex backgrounds and severe class imbalance. (2) Methods. We proposed a semi-automatic labeling pipeline that converts anomaly detection proposals into class labels via small margin cropping from images, interchangeable embeddings (HOG, ResNet-50, ViT-B/16), clustering (k-means/GMM/HDBSCAN), and cluster-level verification using representative montages. (3) Results. On 9354 cropped defects spanning 10 categories (imbalance IR ≈ 1542, Gini ≈ 0.642), ResNet-50 + HDBSCAN achieved NMI ≈ 0.290, AMI ≈ 0.283, and purity ≈ 0.624 with ~47 clusters; ViT + HDBSCAN was comparable (NMI ≈ 0.281, AMI ≈ 0.274, ~44 clusters). With a fixed taxonomy, k-means (K = 10) yielded the strongest ARI (0.169 with ResNet-50; 0.158 with ViT). Macro-purity exceeded micro-purity, indicating many small, homogeneous clusters suitable for one-shot acceptance/rejection, enabling an upper-bound ~200× reduction in operator decisions relative to per-image labeling. (4) Conclusions. The workflow provides an auditable, resource-flexible path from normal-only localization to scalable supervision, prioritizing labeling productivity over detector state-of-the-art and directly addressing the industrial bottleneck in the development lifecycle for PCB inspection.

This novel automated hybrid method can accurately identify landmarks on wrist radiographs and automatically generate the radiological parameters of the distal radius. This method saves time and reduces human labor in creating datasets for training segmentation models and developing image processing algorithms.

The purpose of this study was to directly and quantitatively measure BMD from Cone-beam CT (CBCT) images by enhancing the linearity and uniformity of the bone intensities based on a hybrid deep-learning model (QCBCT-NET) of combining the generative adversarial network (Cycle-GAN) and U-Net, and to compare the bone images enhanced by the QCBCT-NET with those by Cycle-GAN and U-Net. We used two phantoms of human skulls encased in acrylic, one for the training and validation datasets, and the other for the test dataset. We proposed the QCBCT-NET consisting of Cycle-GAN with residual blocks and a multi-channel U-Net using paired training data of quantitative CT (QCT) and CBCT images. The BMD images produced by QCBCT-NET significantly outperformed the images produced by the Cycle-GAN or the U-Net in mean absolute difference (MAD), peak signal to noise ratio (PSNR), normalized cross-correlation (NCC), structural similarity (SSIM), and linearity when compared to the original QCT image. The QCBCT-NET improved the contrast of the bone images by reflecting the original BMD distribution of the QCT image locally using the Cycle-GAN, and also spatial uniformity of the bone images by globally suppressing image artifacts and noise using the two-channel U-Net. The QCBCT-NET substantially enhanced the linearity, uniformity, and contrast as well as the anatomical and quantitative accuracy of the bone images, and demonstrated more accuracy than the Cycle-GAN and the U-Net for quantitatively measuring BMD in CBCT.

We developed an automatic method for staging periodontitis on dental panoramic radiographs using the deep learning hybrid method. A novel hybrid framework was proposed to automatically detect and classify the periodontal bone loss of each individual tooth. The framework is a hybrid of deep learning architecture for detection and conventional CAD processing for classification. Deep learning was used to detect the radiographic bone level (or the CEJ level) as a simple structure for the whole jaw on panoramic radiographs. Next, the percentage rate analysis of the radiographic bone loss combined the tooth long-axis with the periodontal bone and CEJ levels. Using the percentage rate, we could automatically classify the periodontal bone loss. This classification was used for periodontitis staging according to the new criteria proposed at the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. The Pearson correlation coefficient of the automatic method with the diagnoses by radiologists was 0.73 overall for the whole jaw (p < 0.01), and the intraclass correlation value 0.91 overall for the whole jaw (p < 0.01). The novel hybrid framework that combined deep learning architecture and the conventional CAD approach demonstrated high accuracy and excellent reliability in the automatic diagnosis of periodontal bone loss and staging of periodontitis.

The blood glucose prediction algorithms were widely studied for the patients with type 1 diabetes. We analyzed the data of 16 type 2 diabetic patients underwent continuous glucose monitoring (CGM) through deep-learning models. Feed Forward Neural Network (FFNN) and Long Short-Term Memory (LSTM) were used to model and predict the blood glucose. Blood glucose levels after 5, 15, 30, and 45 min were predicted for each patient. To evaluate the predictive power of model, the Root Mean Square Error (RMSE) was calculated by comparing the actual blood glucose and predictive blood glucose. The value of RMSE by FFNN was lower than that by LSTM. The predictive success rates for the glycemic excursion at 5 minutes by FFNN, LSTM were 96% and 95%, the predictive power of two models gradually decreased with time (Figure 1). These results suggest that the well trained deep-learning model would be possible to predict the harmful glycemic events among type 2 diabetic patient. Disclosure D. Kim: None. H. Lee: None. Y. Kim: None. S. Kim: None. S. Lee: None. S. Chun: None.