Search

BackgroundGestational diabetes mellitus (GDM) can cause adverse consequences to both mothers and their newborns. However, pregnant women living in low- and middle-income areas or countries often fail to receive early clinical interventions at local medical facilities due to restricted availability of GDM diagnosis. The outstanding performance of artificial intelligence (AI) in disease diagnosis in previous studies demonstrates its promising applications in GDM diagnosis. ObjectiveThis study aims to investigate the implementation of a well-performing AI algorithm in GDM diagnosis in a setting, which requires fewer medical equipment and staff and to establish an app based on the AI algorithm. This study also explores possible progress if our app is widely used. MethodsAn AI model that included 9 algorithms was trained on 12,304 pregnant outpatients with their consent who received a test for GDM in the obstetrics and gynecology department of the First Affiliated Hospital of Jinan University, a local hospital in South China, between November 2010 and October 2017. GDM was diagnosed according to American Diabetes Association (ADA) 2011 diagnostic criteria. Age and fasting blood glucose were chosen as critical parameters. For validation, we performed k-fold cross-validation (k=5) for the internal dataset and an external validation dataset that included 1655 cases from the Prince of Wales Hospital, the affiliated teaching hospital of the Chinese University of Hong Kong, a non-local hospital. Accuracy, sensitivity, and other criteria were calculated for each algorithm. ResultsThe areas under the receiver operating characteristic curve (AUROC) of external validation dataset for support vector machine (SVM), random forest, AdaBoost, k-nearest neighbors (kNN), naive Bayes (NB), decision tree, logistic regression (LR), eXtreme gradient boosting (XGBoost), and gradient boosting decision tree (GBDT) were 0.780, 0.657, 0.736, 0.669, 0.774, 0.614, 0.769, 0.742, and 0.757, respectively. SVM also retained high performance in other criteria. The specificity for SVM retained 100% in the external validation set with an accuracy of 88.7%. ConclusionsOur prospective and multicenter study is the first clinical study that supports the GDM diagnosis for pregnant women in resource-limited areas, using only fasting blood glucose value, patients’ age, and a smartphone connected to the internet. Our study proved that SVM can achieve accurate diagnosis with less operation cost and higher efficacy. Our study (referred to as GDM-AI study, ie, the study of AI-based diagnosis of GDM) also shows our app has a promising future in improving the quality of maternal health for pregnant women, precision medicine, and long-distance medical care. We recommend future work should expand the dataset scope and replicate the process to validate the performance of the AI algorithms.

Background: diabetices foot ulcer (DFU) are serious complications. It is crucial to detect and diagnose DFU early in order to provide timely treatment, improve patient quality of life, and avoid the social and economic consequences. Machine learning techniques can help identify risk factors associated with DFU development. Objective: The aim of this study was to establish correlations between clinical and biochemical risk factors of DFU through local foot examinations based on the construction of predictive models using automated machine learning techniques. Methods: The input dataset consisted of 566 diabetes cases and 50 DFU risk factors, including 9 local foot examinations. 340 patients with Class 0 labeling (low-risk DFU), 226 patients with Class 1 labeling (high-risk DFU). To divide the training group (consisting of 453 cases) and the validation group (consisting of 113 cases), as well as preprocess the data and develop a prediction model, a Monte Carlo cross-validation approach was employed. Furthermore, potential high-risk factors were analyzed using various algorithms, including Bayesian BYS, Multi-Gaussian Weighted Classifier (MGWC), Support Vector Machine (SVM), and Random Forest Classifier (RF). A three-layer machine learning training was constructed, and model performance was estimated using a Confusion Matrix. The top 30 ranking feature variables were ultimately determined. To reinforce the robustness and generalizability of the predictive model, an independent dataset comprising 248 cases was employed for external validation. This validation process evaluated the model's applicability and reliability across diverse populations and clinical settings. Importantly, the external dataset required no additional tuning or adjustment of parameters, enabling an unbiased assessment of the model's generalizability and its capacity to predict the risk of DFU. Results: The ensemble learning method outperformed individual classifiers in various performance evaluation metrics. Based on the ROC analysis, the AUC of the AutoML model for assessing diabetic foot risk was 88.48 % (74.44–97.83 %). Other results were found to be as follows: 87.23 % (63.33 %–100.00 %) for sensitivity, 87.43 % (70.00 %–100.00 %) for specificity, 87.33 % (76.66 %–95.00 %) for accuracy, 87.69 % (75.00 %–100.00 %) for positive predictive value, and 87.70 % (71.79 %–100.00 %) for negative predictive value. In addition to traditional DFU risk factors such as cardiovascular disorders, peripheral artery disease, and neurological damage, we identified new risk factors such as lower limb varicose veins, history of cerebral infarction, blood urea nitrogen, GFR (Glomerular Filtration Rate), and type of diabetes that may be related to the development of DFU. In the external validation set of 158 samples, originating from an initial 248 with exclusions due to missing labels or features, the model still exhibited strong predictive accuracy. The AUC score of 0.762 indicated a strong discriminatory capability of the model. Furthermore, the Sensitivity and Specificity values provided insights into the model's ability to correctly identify both DFU cases and non-cases, respectively. Conclusion: The predictive model, developed through AutoML and grounded in local foot examinations, has proven to be a robust and practical instrument for the screening, prediction, and diagnosis of DFU risk. This model not only aids medical practitioners in the identification of potential DFU cases but also plays a pivotal role in mitigating the progression towards adverse outcomes. And the recent successful external validation of our DFU risk prediction model marks a crucial advancement, indicating its readiness for clinical application. This validation reinforces the model's efficacy as an accessible and reliable tool for early DFU risk assessment, thereby facilitating prompt intervention strategies and enhancing overall patient outcomes.