You-Cheng Zheng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
You-Cheng Zheng has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Chuan-Sheng Hung
Department of Computer Science and Engineering, National Sun Yat-Sen University, Kaohsiung 804, Taiwan
3 shared publications - Chun-Hung Richard Lin
Department of Computer Science and Engineering, National Sun Yat-Sen University, Kaohsiung 804, Taiwan
3 shared publications - Chih-Min Tsai
Department of Computer Science and Engineering, National Sun Yat-Sen University, Kaohsiung 804, Taiwan
2 shared publications - Ting-Hsin Huang
Department of Computer Science and Engineering, National Sun Yat-Sen University, Kaohsiung 804, Taiwan
2 shared publications - Cheng-Han Yu
Department of Computer Science and Engineering, National Sun Yat-Sen University, Kaohsiung 804, Taiwan
1 shared publication - Cheng-Han Yu
Department of Computer Science and Engineering, National Sun Yat-Sen University, Kaohsiung 804, Taiwan
1 shared publication
Latest Publications
- Open Access
- 30 July 2025
- Published by MDPI AG in Bioengineering
- Vol. 12 (8) , 827
FADEL: Ensemble Learning Enhanced by Feature Augmentation and Discretization
In recent years, data augmentation techniques have become the predominant approach for addressing highly imbalanced classification problems in machine learning. Algorithms such as the Synthetic Minority Over-sampling Technique (SMOTE) and Conditional Tabular Generative Adversarial Network (CTGAN) have proven effective in synthesizing minority class samples. However, these methods often introduce distributional bias and noise, potentially leading to model overfitting, reduced predictive performance, increased computational costs, and elevated cybersecurity risks. To overcome these limitations, we propose a novel architecture, FADEL, which integrates feature-type awareness with a supervised discretization strategy. FADEL introduces a unique feature augmentation ensemble framework that preserves the original data distribution by concurrently processing continuous and discretized features. It dynamically routes these feature sets to their most compatible base models, thereby improving minority class recognition without the need for data-level balancing or augmentation techniques. Experimental results demonstrate that FADEL, solely leveraging feature augmentation without any data augmentation, achieves a recall of 90.8% and a G-mean of 94.5% on the internal test set from Kaohsiung Chang Gung Memorial Hospital in Taiwan. On the external validation set from Kaohsiung Medical University Chung-Ho Memorial Hospital, it maintains a recall of 91.9% and a G-mean of 86.7%. These results outperform conventional ensemble methods trained on CTGAN-balanced datasets, confirming the superior stability, computational efficiency, and cross-institutional generalizability of the FADEL architecture. Altogether, FADEL uses feature augmentation to offer a robust and practical solution to extreme class imbalance, outperforming mainstream data augmentation-based approaches.
- 30 July 2025
- Published by MDPI AG in Bioengineering
- Vol. 12 (8) , 827
Open AccessFADEL: Ensemble Learning Enhanced by Feature Augmentation and Discretization
In recent years, data augmentation techniques have become the predominant approach for addressing highly imbalanced classification problems in machine learning. Algorithms such as the Synthetic Minority Over-sampling Technique (SMOTE) and Conditional Tabular Generative Adversarial Network (CTGAN) have proven effective in synthesizing minority class samples. However, these methods often introduce distributional bias and noise, potentially leading to model overfitting, reduced predictive performance, increased computational costs, and elevated cybersecurity risks. To overcome these limitations, we propose a novel architecture, FADEL, which integrates feature-type awareness with a supervised discretization strategy. FADEL introduces a unique feature augmentation ensemble framework that preserves the original data distribution by concurrently processing continuous and discretized features. It dynamically routes these feature sets to their most compatible base models, thereby improving minority class recognition without the need for data-level balancing or augmentation techniques. Experimental results demonstrate that FADEL, solely leveraging feature augmentation without any data augmentation, achieves a recall of 90.8% and a G-mean of 94.5% on the internal test set from Kaohsiung Chang Gung Memorial Hospital in Taiwan. On the external validation set from Kaohsiung Medical University Chung-Ho Memorial Hospital, it maintains a recall of 91.9% and a G-mean of 86.7%. These results outperform conventional ensemble methods trained on CTGAN-balanced datasets, confirming the superior stability, computational efficiency, and cross-institutional generalizability of the FADEL architecture. Altogether, FADEL uses feature augmentation to offer a robust and practical solution to extreme class imbalance, outperforming mainstream data augmentation-based approaches.
- Open Access
- 3 July 2025
- Published by MDPI AG in Bioengineering
- Vol. 12 (7) , 730
Predicting Fetal Growth with Curve Fitting and Machine Learning
Monitoring fetal growth throughout pregnancy is essential for early detection of developmental abnormalities. This study developed a Taiwan-specific fetal growth reference using a web-based data collection platform and polynomial regression modeling. We analyzed ultrasound data from 980 pregnant women, encompassing 8350 prenatal scans, to model six key fetal biometric parameters: abdominal circumference, crown–rump length, estimated fetal weight, head circumference, biparietal diameter, and femur length. Quadratic regression was selected based on a balance of performance and simplicity, with R2 values exceeding 0.95 for most parameters. Confidence intervals and real-time anomaly detection were implemented through the platform. The results demonstrate the potential for efficient, population-specific fetal growth monitoring in clinical settings.
- 3 July 2025
- Published by MDPI AG in Bioengineering
- Vol. 12 (7) , 730
Open AccessPredicting Fetal Growth with Curve Fitting and Machine Learning
Monitoring fetal growth throughout pregnancy is essential for early detection of developmental abnormalities. This study developed a Taiwan-specific fetal growth reference using a web-based data collection platform and polynomial regression modeling. We analyzed ultrasound data from 980 pregnant women, encompassing 8350 prenatal scans, to model six key fetal biometric parameters: abdominal circumference, crown–rump length, estimated fetal weight, head circumference, biparietal diameter, and femur length. Quadratic regression was selected based on a balance of performance and simplicity, with R2 values exceeding 0.95 for most parameters. Confidence intervals and real-time anomaly detection were implemented through the platform. The results demonstrate the potential for efficient, population-specific fetal growth monitoring in clinical settings.
- Open Access
- 31 March 2024
- Published by MDPI AG in Bioengineering
- Vol. 11 (4) , 345
Utilizing Nearest-Neighbor Clustering for Addressing Imbalanced Datasets in Bioengineering
Imbalance classification is common in scenarios like fault diagnosis, intrusion detection, and medical diagnosis, where obtaining abnormal data is difficult. This article addresses a one-class problem, implementing and refining the One-Class Nearest-Neighbor (OCNN) algorithm. The original inter-quartile range mechanism is replaced with the K-means with outlier removal (KMOR) algorithm for efficient outlier identification in the target class. Parameters are optimized by treating these outliers as non-target-class samples. A new algorithm, the Location-based Nearest-Neighbor (LBNN) algorithm, clusters one-class training data using KMOR and calculates the farthest distance and percentile for each test data point to determine if it belongs to the target class. Experiments cover parameter studies, validation on eight standard imbalanced datasets from KEEL, and three applications on real medical imbalanced datasets. Results show superior performance in precision, recall, and G-means compared to traditional classification models, making it effective for handling imbalanced data challenges.
- 31 March 2024
- Published by MDPI AG in Bioengineering
- Vol. 11 (4) , 345
Open AccessUtilizing Nearest-Neighbor Clustering for Addressing Imbalanced Datasets in Bioengineering
Imbalance classification is common in scenarios like fault diagnosis, intrusion detection, and medical diagnosis, where obtaining abnormal data is difficult. This article addresses a one-class problem, implementing and refining the One-Class Nearest-Neighbor (OCNN) algorithm. The original inter-quartile range mechanism is replaced with the K-means with outlier removal (KMOR) algorithm for efficient outlier identification in the target class. Parameters are optimized by treating these outliers as non-target-class samples. A new algorithm, the Location-based Nearest-Neighbor (LBNN) algorithm, clusters one-class training data using KMOR and calculates the farthest distance and percentile for each test data point to determine if it belongs to the target class. Experiments cover parameter studies, validation on eight standard imbalanced datasets from KEEL, and three applications on real medical imbalanced datasets. Results show superior performance in precision, recall, and G-means compared to traditional classification models, making it effective for handling imbalanced data challenges.
- Open Access
Prospective Clinical Evaluation of a Deep Learning Algorithm for Guided Point-of-Care Ultrasonography Screening of Abdominal Aortic Aneurysms
Background Abdominal Aortic Aneurysm (AAA) is a critical condition that can lead to fatal consequences if not detected and treated early. Despite the high prevalence in smokers and guideline recommendation for screening, AAA often remains undetected due to availability of diagnostic ultrasound examinations. This prospective clinical trial aimed to investigate the use of a Deep Learning (DL) algorithm to guide AAA screening. Methods This prospective, comparative diagnostic study was conducted at the Kaohsiung Chang Gung Memorial Hospital. We developed and deployed an object detection-based DL algorithm providing real-time guidance for novice users performing AAA screening using point of care ultrasound. 10 registered nurses with no prior ultrasonography experience were recruited and performed at least 15 scans on patients over 65 years old to acquire abdominal aorta videos. These scans were compared with those of physicians using the same ultrasound hardware but without DL guidance. Results A total of 184 patients (median [IQR] age of 72 [67-79], and 105 (57.1%) male) completed this study. The DL-guided novices achieved adequate scan quality in 87.5% (95% CI: 82.7 - 92.3%) of patients, comparable to the 91.3% (95% CI: 87.2-95.4%) rate of physician scans (p=0.310). This performance did not vary by BMI. The DL model predicted AAA with an AUC of 0.975, showing 100% sensitivity and 94.3% specificity. The DL model predicted the maximal width of abdominal aorta with mean absolute error of 2.8mm compared to physician measurements. 3 AAA with maximal width of aorta > 3cm were found in this study cohort. Conclusion DL-guided POCUS is an effective tool for AAA screening, providing comparable performance to experienced physicians. The use of this DL system could democratize AAA screening and improve access, thereby aiding in early disease detection and treatment. What is New Our study presents a deep learning (DL) guidance system that enables novice users to perform Abdominal Aortic Aneurysm (AAA) screening with POCUS, yielding image quality comparable to experienced physicians. The DL algorithm accurately identifies AAA from scans conducted by novice users, maintains consistent performance across patients with varying BMIs, and demonstrates increased scan efficiency with repeated use. Clinical Implications DL-guided POCUS can potentially expand AAA screening capabilities to non-specialist settings and increase throughput for screening at risk individuals. The implementation of our DL model for AAA screening could enhance early detection, particularly in underserved areas, but also optimize clinical workflows by decreasing diagnostic wait times and increasing ultrasound utilization efficiency.
Open AccessProspective Clinical Evaluation of a Deep Learning Algorithm for Guided Point-of-Care Ultrasonography Screening of Abdominal Aortic Aneurysms
Background Abdominal Aortic Aneurysm (AAA) is a critical condition that can lead to fatal consequences if not detected and treated early. Despite the high prevalence in smokers and guideline recommendation for screening, AAA often remains undetected due to availability of diagnostic ultrasound examinations. This prospective clinical trial aimed to investigate the use of a Deep Learning (DL) algorithm to guide AAA screening. Methods This prospective, comparative diagnostic study was conducted at the Kaohsiung Chang Gung Memorial Hospital. We developed and deployed an object detection-based DL algorithm providing real-time guidance for novice users performing AAA screening using point of care ultrasound. 10 registered nurses with no prior ultrasonography experience were recruited and performed at least 15 scans on patients over 65 years old to acquire abdominal aorta videos. These scans were compared with those of physicians using the same ultrasound hardware but without DL guidance. Results A total of 184 patients (median [IQR] age of 72 [67-79], and 105 (57.1%) male) completed this study. The DL-guided novices achieved adequate scan quality in 87.5% (95% CI: 82.7 - 92.3%) of patients, comparable to the 91.3% (95% CI: 87.2-95.4%) rate of physician scans (p=0.310). This performance did not vary by BMI. The DL model predicted AAA with an AUC of 0.975, showing 100% sensitivity and 94.3% specificity. The DL model predicted the maximal width of abdominal aorta with mean absolute error of 2.8mm compared to physician measurements. 3 AAA with maximal width of aorta > 3cm were found in this study cohort. Conclusion DL-guided POCUS is an effective tool for AAA screening, providing comparable performance to experienced physicians. The use of this DL system could democratize AAA screening and improve access, thereby aiding in early disease detection and treatment. What is New Our study presents a deep learning (DL) guidance system that enables novice users to perform Abdominal Aortic Aneurysm (AAA) screening with POCUS, yielding image quality comparable to experienced physicians. The DL algorithm accurately identifies AAA from scans conducted by novice users, maintains consistent performance across patients with varying BMIs, and demonstrates increased scan efficiency with repeated use. Clinical Implications DL-guided POCUS can potentially expand AAA screening capabilities to non-specialist settings and increase throughput for screening at risk individuals. The implementation of our DL model for AAA screening could enhance early detection, particularly in underserved areas, but also optimize clinical workflows by decreasing diagnostic wait times and increasing ultrasound utilization efficiency.