A feature optimization study based on a diabetes risk questionnaire.

Introduction: The prevalence of diabetes, a common chronic disease, has shown a gradual increase, posing substantial burdens on both society and individuals. In order to enhance the effectiveness of diabetes risk prediction questionnaires, optimize the selection of characteristic variables, and raise awareness of diabetes risk among residents, this study utilizes survey data obtained from the risk factor monitoring system of the Centers for Disease Control and Prevention in the United States. Methods: Following univariate analysis and meticulous screening, a more refined dataset was constructed. This dataset underwent preprocessing steps, including data distribution standardization, the application of the Synthetic Minority Oversampling Technique (SMOTE) in combination with the Round function for equilibration, and data standardization. Subsequently, machine learning (ML) techniques were employed, utilizing enumerated feature variables to evaluate the strength of the correlation among diabetes risk factors. Results: The research findings effectively delineated the ranking of characteristic variables that significantly influence the risk of diabetes. Obesity emerges as the most impactful factor, overshadowing other risk factors. Additionally, psychological factors, advanced age, high cholesterol, high blood pressure, alcohol abuse, coronary heart disease or myocardial infarction, mobility difficulties, and low family income exhibit correlations with diabetes risk to varying degrees. Discussion: The experimental data in this study illustrate that, while maintaining comparable accuracy, optimization of questionnaire variables and the number of questions can significantly enhance efficiency for subsequent follow-up and precise diabetes prevention. Moreover, the research methods employed in this study offer valuable insights into studying the risk correlation of other diseases, while the research results contribute to heightened societal awareness of populations at elevated risk of diabetes.

Diabetes prediction model for unbalanced community follow-up data set based on optimal feature selection and scorecard.

Objectives: Diabetes is a metabolic disease and early detection is crucial to ensuring a healthy life for people with prediabetes. Community care plays an important role in public health, but the association between community follow-up of key life characteristics and diabetes risk remains unclear. Based on the method of optimal feature selection and risk scorecard, follow-up data of diabetes patients are modeled to assess diabetes risk. Methods: We conducted a study on the diabetes risk assessment model and risk scorecard using follow-up data from diabetes patients in Haizhu District, Guangzhou, from 2016 to 2023. The raw data underwent preprocessing and imbalance handling. Subsequently, features relevant to diabetes were selected and optimized to determine the optimal subset of features associated with community follow-up and diabetes risk. We established the diabetes risk assessment model. Furthermore, for a comprehensible and interpretable risk expression, the Weight of Evidence transformation method was applied to features. The transformed features were discretized using the quantile binning method to design the risk scorecard, mapping the model's output to five risk levels. Results: In constructing the diabetes risk assessment model, the Random Forest classifier achieved the highest accuracy. The risk scorecard obtained an accuracy of 85.16%, precision of 87.30%, recall of 80.26%, and an F1 score of 83.27% on the unbalanced research dataset. The performance loss compared to the diabetes risk assessment model was minimal, suggesting that the binning method used for constructing the diabetes risk scorecard is reasonable, with very low feature information loss. Conclusion: The methods provided in this article demonstrate effectiveness and reliability in the assessment of diabetes risk. The assessment model and scorecard can be directly applied to community doctors for large-scale risk identification and early warning and can also be used for individual self-examination to reduce risk factor levels.

Diabetes risk prediction model based on community follow-up data using machine learning.

Diabetes is a chronic metabolic disease characterized by hyperglycemia, the follow-up management of diabetes patients is mostly in the community, but the relationship between key lifestyle indicators in community follow-up and the risk of diabetes is unclear. In order to explore the association between key life characteristic indicators of community follow-up and the risk of diabetes, 252,176 follow-up records of people with diabetes patients from 2016 to 2023 were obtained from Haizhu District, Guangzhou. According to the follow-up data, the key life characteristic indicators that affect diabetes are determined, and the optimal feature subset is obtained through feature selection technology to accurately assess the risk of diabetes. A diabetes risk assessment model based on a random forest classifier was designed, which used optimal feature parameter selection and algorithm model comparison, with an accuracy of 91.24% and an AUC corresponding to the ROC curve of 97%. In order to improve the applicability of the model in clinical and real life, a diabetes risk score card was designed and tested using the original data, the accuracy was 95.15%, and the model reliability was high. The diabetes risk prediction model based on community follow-up big data mining can be used for large-scale risk screening and early warning by community doctors based on patient follow-up data, further promoting diabetes prevention and control strategies, and can also be used for wearable devices or intelligent biosensors for individual patient self examination, in order to improve lifestyle and reduce risk factor levels.

Recover User's Private Training Image Data by Gradient in Federated Learning.

Exchanging gradient is a widely used method in modern multinode machine learning system (e.g., distributed training, Federated Learning). Gradients and weights of model has been presumed to be safe to delivery. However, some studies have shown that gradient inversion technique can reconstruct the input images on the pixel level. In this study, we review the research work of data leakage by gradient inversion technique and categorize existing works into three groups: (i) Bias Attacks, (ii) Optimization-Based Attacks, and (iii) Linear Equation Solver Attacks. According to the characteristics of these algorithms, we propose one privacy attack system, i.e., Single-Sample Reconstruction Attack System (SSRAS). This system can carry out image reconstruction regardless of whether the label can be determined. It can extends gradient inversion attack from a fully connected layer with bias terms to attack a fully connected layer and convolutional neural network with or without bias terms. We also propose Improved R-GAP Alogrithm, which can utlize DLG algorithm to derive ground truth. Furthermore, we introduce Rank Analysis Index (RA-I) to measure the possible of whether the user's raw image data can be reconstructed. This rank analysis derive virtual constraints Vi from weights. Compared with the most representative attack algorithms, this reconstruction attack system can recover a user's private training image with high fidelity and attack success rate. Experimental results also show the superiority of the attack system over some other state-of-the-art attack algorithms.

Broadband high quantum efficiency InGaAs/InP focal plane arrays via high precision plasma thinning.

We report on the fabrication of a 160×120 visible (Vis)-extended InGaAs/InP focal plane array (FPA) by means of the inductively coupled plasma etching. Compared to the conventional Vis-InGaAs FPAs, higher quantum efficiency in the Vis spectrum has been achieved. High precision thinning of the n-type InP contact layer down to 10 nm has led to quantum efficiencies higher than 60% over a broad wavelength range of 500-1700 nm. Benefiting from the textured surface after plasma etching, 17% lower reflectance over the entire response range was also found. Enhanced Vis/near-infrared (NIR) laboratory imaging capability has also been demonstrated, which has proved the feasibility of such processes for fabrication of future higher definition Vis/NIR InGaAs imagers.

IGA-rule 17 for performance estimation of wavelength-extended InGaAs photodetectors: validity and limitations.

A rule of thumb, denoted as IGA-rule 17, has been developed to describe the temperature and cutoff wavelength-dependent dark currents of wavelength-extended InGaAs photodetectors in a 2-3 µm band. The validity and limitations of the rule are discussed. This rule is intended as an index for device developers to evaluate their technologies in processing, a simple tool for device users to estimate reachable performance at various conditions in their design, and an effective bridge between the two.

Nanoscale imaging of the photoresponse in PN junctions of InGaAs infrared detector.

Electronic layout, such as distributions of charge carriers and electric field, in PN junction is determinant for the photovoltaic devices to realize their functionality. Considerable efforts have been dedicated to the carrier profiling of this specific region with Scanning Probe Microscope, yet reliable analysis was impeded by the difficulty in resolving carriers with high mobility and the unclear surface effect, particularly on compound semiconductors. Here we realize nanometer Scanning Capacitance Microscopic study on the cross-section of InGaAs/InP photodetectors with the featured dC/dV layout of PN junction unveiled for the first time. It enables us to probe the photo-excited minority carriers in junction region and diagnose the performance deficiency of the diode devices. This work provides an illuminating insight into the PN junction for assessing its basic capability of harvesting photo-carriers as well as blocking leakage current in nanoscopic scale.

Subwavelength Gold Grating as Polarizers Integrated with InP-Based InGaAs Sensors.

There are currently growing needs for polarimetric imaging in infrared wavelengths for broad applications in bioscience, communications and agriculture, etc. Subwavelength metallic gratings are capable of separating transverse magnetic (TM) mode from transverse electric (TE) mode to form polarized light, offering a reliable approach for the detection in polarization way. This work aims to design and fabricate subwavelength gold gratings as polarizers for InP-based InGaAs sensors in 1.0-1.6 µm. The polarization capability of gold gratings on InP substrate with pitches in the range of 200-1200 nm (fixed duty cycle of 0.5) has been systematically studied by both theoretical modeling with a finite-difference time-domain (FDTD) simulator and spectral measurements. Gratings with 200 nm lines/space in 100-nm-thick gold have been fabricated by electron beam lithography (EBL). It was found that subwavelength gold gratings directly integrated on InP cannot be applied as good polarizers, because of the existence of SPP modes in the detection wavelengths. An effective solution has been found by sandwiching the Au/InP bilayer using a 200 nm SiO2 layer, leading to significant improvement in both TM transmission and extinction ratio. At 1.35 µm, the improvement factors are 8 and 10, respectively. Therefore, it is concluded that the Au/SiO2/InP trilayer should be a promising candidate of near-infrared polarizers for the InP-based InGaAs sensors.