Prediction of OCT images of short-term response to anti-VEGF treatment for diabetic macular edema using different generative adversarial networks

Prediction of Long-Term Treatment Outcomes for Diabetic Macular Edema Using a Generative Adversarial Network

Purpose

The purpose of this study was to analyze optical coherence tomography (OCT) images of generative adversarial networks (GANs) for the prediction of diabetic macular edema after long-term treatment.

Methods

Diabetic macular edema (DME) eyes (n = 327) underwent anti-vascular endothelial growth factor (VEGF) treatments every 4 weeks for 52 weeks from a randomized controlled trial (CRTH258B2305, KINGFISHER) were included. OCT B-scan images through the foveal center at weeks 0, 4, 12, and 52, fundus photography, and retinal thickness (RT) maps were collected. GAN models were trained to generate probable OCT images after treatment. Input for each model were comprised of either the baseline B-scan alone or combined with additional OCT, thickness map, or fundus images. Generated OCT B-scan images were compared with real week 52 images.

Results

For 30 test images, 28, 29, 15, and 30 gradable OCT images were generated by CycleGAN, UNIT, Pix2PixHD, and RegGAN, respectively. In comparison with the real week 52, these GAN models showed positive predictive value (PPV), sensitivity, specificity, and kappa for residual fluid ranging from 0.500 to 0.889, 0.455 to 1.000, 0.357 to 0.857, and 0.537 to 0.929, respectively. For hard exudate (HE), they were ranging from 0.500 to 1.000, 0.545 to 0.900, 0.600 to 1.000, and 0.642 to 0.894, respectively. Models trained with week 4 and 12 B-scans as additional inputs to the baseline B-scan showed improved performance.

Conclusions

GAN models could predict residual fluid and HE after long-term anti-VEGF treatment of DME.

Translational Relevance

The implementation of this tool may help identify potential nonresponders after long-term treatment, thereby facilitating management planning for these eyes.

Generative deep learning for the development of a type 1 diabetes simulator

BACKGROUND

Type 1 diabetes (T1D) simulators, crucial for advancing diabetes treatments, often fall short of capturing the entire complexity of the glucose-insulin system due to the imprecise approximation of the physiological models. This study introduces a simulation approach employing a conditional deep generative model. The aim is to overcome the limitations of existing T1D simulators by synthesizing virtual patients that more accurately represent the entire glucose-insulin system physiology.

METHODS

Our methodology utilizes a sequence-to-sequence generative adversarial network to simulate virtual T1D patients causally. Causality is embedded in the model by introducing shifted input-output pairs during training, with a 90-min shift capturing the impact of input insulin and carbohydrates on blood glucose. To validate our approach, we train and evaluate the model using three distinct datasets, each consisting of 27, 12, and 10 T1D patients, respectively. In addition, we subject the trained model to further validation for closed-loop therapy, employing a state-of-the-art controller.

RESULTS

The generated patients display statistical similarity to real patients when evaluated on the time-in-range results for each of the standard blood glucose ranges in T1D management along with means and variability outcomes. When tested for causality, authentic causal links are identified between the insulin, carbohydrates, and blood glucose levels of the virtual patients. The trained generative model demonstrates behaviours that are closer to reality compared to conventional T1D simulators when subjected to closed-loop insulin therapy using a state-of-the-art controller.

CONCLUSIONS

These results highlight our approach's capability to accurately capture physiological dynamics and establish genuine causal relationships, holding promise for enhancing the development and evaluation of therapies in diabetes.

Choice of refractive surgery types for myopia assisted by machine learning based on doctors' surgical selection data

In recent years, corneal refractive surgery has been widely used in clinics as an effective means to restore vision and improve the quality of life. When choosing myopia-refractive surgery, it is necessary to comprehensively consider the differences in equipment and technology as well as the specificity of individual patients, which heavily depend on the experience of ophthalmologists. In our study, we took advantage of machine learning to learn about the experience of ophthalmologists in decision-making and assist them in the choice of corneal refractive surgery in a new case. Our study was based on the clinical data of 7,081 patients who underwent corneal refractive surgery between 2000 and 2017 at the Department of Ophthalmology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences. Due to the long data period, there were data losses and errors in this dataset. First, we cleaned the data and deleted the samples of key data loss. Then, patients were divided into three groups according to the type of surgery, after which we used SMOTE technology to eliminate imbalance between groups. Six statistical machine learning models, including NBM, RF, AdaBoost, XGBoost, BP neural network, and DBN were selected, and a ten-fold cross-validation and grid search were used to determine the optimal hyperparameters for better performance. When tested on the dataset, the multi-class RF model showed the best performance, with agreement with ophthalmologist decisions as high as 0.8775 and Macro F1 as high as 0.8019. Furthermore, the results of the feature importance analysis based on the SHAP technique were consistent with an ophthalmologist's practical experience. Our research will assist ophthalmologists in choosing appropriate types of refractive surgery and will have beneficial clinical effects.