Dataset Information

Generative adversarial networks for imputing missing data for big data clinical research.

ABSTRACT:

Background

Missing data is a pervasive problem in clinical research. Generative adversarial imputation nets (GAIN), a novel machine learning data imputation approach, has the potential to substitute missing data accurately and efficiently but has not yet been evaluated in empirical big clinical datasets.

Objectives

This study aimed to evaluate the accuracy of GAIN in imputing missing values in large real-world clinical datasets with mixed-type variables. The computation efficiency of GAIN was also evaluated. The performance of GAIN was compared with other commonly used methods, MICE and missForest.

Methods

Two real world clinical datasets were used. The first was that of a cohort study on the long-term outcomes of patients with diabetes (50,000 complete cases), and the second was of a cohort study on the effectiveness of a risk assessment and management programme for patients with hypertension (10,000 complete cases). Missing data (missing at random) to independent variables were simulated at different missingness rates (20, 50%). The normalized root mean square error (NRMSE) between imputed values and real values for continuous variables and the proportion of falsely classified (PFC) for categorical variables were used to measure imputation accuracy. Computation time per imputation for each method was recorded. The differences in accuracy of different imputation methods were compared using ANOVA or non-parametric test.

Results

Both missForest and GAIN were more accurate than MICE. GAIN showed similar accuracy as missForest when the simulated missingness rate was 20%, but was more accurate when the simulated missingness rate was 50%. GAIN was the most accurate for the imputation of skewed continuous and imbalanced categorical variables at both missingness rates. GAIN had a much higher computation speed (32 min on PC) comparing to that of missForest (1300 min) when the sample size is 50,000.

Conclusion

GAIN showed better accuracy as an imputation method for missing data in large real-world clinical datasets compared to MICE and missForest, and was more resistant to high missingness rate (50%). The high computation speed is an added advantage of GAIN in big clinical data research. It holds potential as an accurate and efficient method for missing data imputation in future big data clinical research.

Trial registration

ClinicalTrials.gov ID: NCT03299010 ; Unique Protocol ID: HKUCTR-2232.

SUBMITTER: Dong W

PROVIDER: S-EPMC8059005 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:PurposeComplementary information obtained from multiple contrasts of tissue facilitates physicians assessing, diagnosing and planning treatment of a variety of diseases. However, acquiring multiple contrasts magnetic resonance images (MRI) for every patient using multiple pulse sequences is time-consuming and expensive, where, medical image synthesis has been demonstrated as an effective alternative. The purpose of this study is to develop a unified framework for multimodal MR image synthesis.MethodsA unified generative adversarial network consisting of only a single generator and a single discriminator was developed to learn the mappings among images of four different modalities. The generator took an image and its modality label as inputs and learned to synthesize the image in the target modality, while the discriminator was trained to distinguish between real and synthesized images and classify them to their corresponding modalities. The network was trained and tested using multimodal brain MRI consisting of four different contrasts which are T1-weighted (T1), T1-weighted and contrast-enhanced (T1c), T2-weighted (T2), and fluid-attenuated inversion recovery (Flair). Quantitative assessments of our proposed method were made through computing normalized mean absolute error (NMAE), peak signal-to-noise ratio (PSNR), structural similarity index measurement (SSIM), visual information fidelity (VIF), and naturalness image quality evaluator (NIQE).ResultsThe proposed model was trained and tested on a cohort of 274 glioma patients with well-aligned multi-types of MRI scans. After the model was trained, tests were conducted by using each of T1, T1c, T2, Flair as a single input modality to generate its respective rest modalities. Our proposed method shows high accuracy and robustness for image synthesis with arbitrary MRI modality that is available in the database as input. For example, with T1 as input modality, the NMAEs for the generated T1c, T2, Flair respectively are 0.034 ± 0.005, 0.041 ± 0.006, and 0.041 ± 0.006, the PSNRs respectively are 32.353 ± 2.525 dB, 30.016 ± 2.577 dB, and 29.091 ± 2.795 dB, the SSIMs are 0.974 ± 0.059, 0.969 ± 0.059, and 0.959 ± 0.059, the VIF are 0.750 ± 0.087, 0.706 ± 0.097, and 0.654 ± 0.062, and NIQE are 1.396 ± 0.401, 1.511 ± 0.460, and 1.259 ± 0.358, respectively.ConclusionsWe proposed a novel multimodal MR image synthesis method based on a unified generative adversarial network. The network takes an image and its modality label as inputs and synthesizes multimodal images in a single forward pass. The results demonstrate that the proposed method is able to accurately synthesize multimodal MR images from a single MR image.