A comparison of the performances of an artificial neural network and a regression model for GFR estimation

Artificial neural network for the prediction model of glomerular filtration rate to estimate the normal or abnormal stages of kidney using gamma camera

OBJECTIVE

Chronic kidney disease (CKD) is evaluated based on glomerular filtration rate (GFR) using a gamma camera in the nuclear medicine center or hospital in a routine procedure, but the gamma camera does not provide the accurate stages of the diseases. Therefore, this research aimed to find out the normal or abnormal stages of CKD based on the value of GFR using an artificial neural network (ANN).

METHODS

Two hundred fifty (Training 188, Testing 62) kidney patients who underwent the ultrasonography test to diagnose the renal test in our nuclear medical centre were scanned using gamma camera. The patients were injected with 99mTc-DTPA before the scanning procedure. After pushing the syringe into the patient's vein, the pre-syringe and post syringe radioactive counts were calculated using the gamma camera. The artificial neural network uses the softmax function with cross-entropy loss to diagnose CKD normal or abnormal labels depending on the value of GFR in the output layer.

RESULTS

The results showed that the accuracy of the proposed ANN model was 99.20% for K-fold cross-validation. The sensitivity and specificity were 99.10% and 99.20%, respectively. The Area under the curve (AUC) was 0.9994.

CONCLUSION

The proposed model using an artificial neural network can classify the normal or abnormal stages of CKD. After implementing the proposed model clinically, it may upgrade the gamma camera to diagnose the normal or abnormal stages of the CKD with an appropriate GFR value.

Artificial Intelligence Enhances Studies on Inflammatory Bowel Disease

Inflammatory bowel disease (IBD), which includes ulcerative colitis (UC) and Crohn’s disease (CD), is an idiopathic condition related to a dysregulated immune response to commensal intestinal microflora in a genetically susceptible host. As a global disease, the morbidity of IBD reached a rate of 84.3 per 100,000 persons and reflected a continued gradual upward trajectory. The medical cost of IBD is also notably extremely high. For example, in Europe, it has €3,500 in CD and €2,000 in UC per patient per year, respectively. In addition, taking into account the work productivity loss and the reduced quality of life, the indirect costs are incalculable. In modern times, the diagnosis of IBD is still a subjective judgment based on laboratory tests and medical images. Its early diagnosis and intervention is therefore a challenging goal and also the key to control its progression. Artificial intelligence (AI)-assisted diagnosis and prognosis prediction has proven effective in many fields including gastroenterology. In this study, support vector machines were utilized to distinguish the significant features in IBD. As a result, the reliability of IBD diagnosis due to its impressive performance in classifying and addressing region problems was improved. Convolutional neural networks are advanced image processing algorithms that are currently in existence. Digestive endoscopic images can therefore be better understood by automatically detecting and classifying lesions. This study aims to summarize AI application in the area of IBD, objectively evaluate the performance of these methods, and ultimately understand the algorithm–dataset combination in the studies.

Improving glomerular filtration rate estimation by semi-supervised learning: a development and external validation study

BACKGROUND

Accurate estimating glomerular filtration rate (GFR) is crucial both in clinical practice and epidemiological survey. We incorporated semi-supervised learning technology to improve GFR estimation performance.

METHODS

AASK [African American Study of Kidney Disease and Hypertension], CRIC [Chronic Renal Insufficiency Cohort] and DCCT [Diabetes Control and Complications Trial] studies were pooled together for model development, whereas MDRD [Modification of Diet in Renal Disease] and CRISP [Consortium for Radiological Imaging Studies of Polycystic Kidney Disease] studies for model external validation. A total of seven variables (Serum creatinine, Age, Sex, Black race, Diabetes status, Hypertension and Body Mass Index) were included as independent variables, while the outcome variable GFR was measured as the urinary clearance of ¹²⁵I-iothalamate. The revised CKD-EPI [Chronic Kidney Disease Epidemiology Collaboration] creatinine equations was selected as benchmark for performance comparisons. Head-to-head performance comparisons from four-variable to seven-variable combination were conducted between revised CKD-EPI equations and semi-supervised models.

RESULTS

In each independent variables combination, the semi-supervised models consistently achieved superior results in all three performance indicators compared with corresponding revised CKD-EPI equations in the external validation data set. Furthermore, compared with revised four-variable CKD-EPI equation, the seven-variable semi-supervised model performed less biased (mean of difference: 0.03 [- 0.28, 0.34] vs 1.53 [1.28, 1.85], P < 0.001), more precise (interquartile range of difference: 7.94 [7.37, 8.50] vs 8.28 [7.76, 8.83], P = 0.1) and accurate (P30: 88.9% [87.4%, 90.2%] vs 86.0% [84.4%, 87.4%], P < 0.001.

CONCLUSIONS

The superior performance of the semi-supervised models during head-to-head comparisons supported the hypothesis that semi-supervised learning technology could improve GFR estimation performance.

Age- and Sex-Specific Reference Values of Estimated Glomerular Filtration Rate in Chinese Population

BACKGROUND

Due to an aging population, prevalence and mortality of CKD continue to increase. Current CKD definition has been challenged recently. Age- and sex-specific reference values of estimated glomerular filtration rate (eGFR) in China are still lacking.

METHODS

Age- and sex-stratified, randomly selected inhabitants received a health examination and an inquest into medical history. The GFR was estimated using CKD-EPI equation. We calculated means with ±1.96 times of standard deviation and 2.5th, 97.5th percentiles of eGFR per 5-year age-group. Some of their GFRs were measured by the Gates method (99mTc-DTPA renal scintigraphy) and estimated by cystatin C-based equation.

RESULTS

The cohort study included 17,037 male and 9,304 female Chinese persons aged 18-99 years. A reference population of apparently healthy subjects was selected by excluding persons with known hypertension, diabetes, cardiovascular, or renal diseases. This healthy cohort study included 12,231 male subjects and 6,463 female subjects. The mean eGFR was higher in the female than that in the male who were younger than 40-year (122 mL/min/1.73 m2 vs. 111 mL/min/1.73 m2). In these apparently healthy persons, GFR declined approximately 0.8 mL/min/year. The lower limit of eGFR (2.5th percentile or mean minus 1.96 times of standard deviation) was <60 or 45 mL/min/1.73 m2 at the age of ≥40 or 65 years old, respectively.

CONCLUSION

The mean eGFR was higher in young females. GFR declined approximately 0.8 mL/min/year. The lower bound of eGFR was <60 mL/min/1.73 m2 or 45 mL/min/1.73 m2 at the age of ≥40 or ≥65 years, respectively. Our study provides age- and sex-specific reference values of GFR in a Chinese population.

Improving accuracy of estimating glomerular filtration rate using artificial neural network: model development and validation

BACKGROUND

The performance of previously published glomerular filtration rate (GFR) estimation equations degrades when directly used in Chinese population. We incorporated more independent variables and using complicated non-linear modeling technology (artificial neural network, ANN) to develop a more accurate GFR estimation model for Chinese population.

METHODS

The enrolled participants came from the Third Affiliated Hospital of Sun Yat-sen University, China from Jan 2012 to Jun 2016. Participants with age < 18, unstable kidney function, taking trimethoprim or cimetidine, or receiving dialysis were excluded. Among the finally enrolled 1952 participants, 1075 participants (55.07%) from Jan 2012 to Dec 2014 were assigned as the development data whereas 877 participants (44.93%) from Jan 2015 to Jun 2016 as the internal validation data. We in total developed 3 GFR estimation models: a 4-variable revised CKD-EPI (chronic kidney disease epidemiology collaboration) equation (standardized serum creatinine and cystatin C, age and gender), a 9-variable revised CKD-EPI equation (additional auxiliary variables: body mass index, blood urea nitrogen, albumin, uric acid and hemoglobin), and a 9-variable ANN model.

RESULTS

Compared with the 4-variable equation, the 9-variable equation could not achieve superior performance in the internal validation data (mean of difference: 5.00 [3.82, 6.54] vs 4.67 [3.55, 5.90], P = 0.5; interquartile range (IQR) of difference: 18.91 [17.43, 20.48] vs 20.11 [18.46, 21.80], P = 0.05; P30: 76.6% [73.7%, 79.5%] vs 75.8% [72.9%, 78.6%], P = 0.4), but the 9-variable ANN model significantly improve bias and P30 accuracy (mean of difference: 2.77 [1.82, 4.10], P = 0.007; IQR: 19.33 [17.77, 21.17], P = 0.3; P30: 80.0% [77.4%, 82.7%], P < 0.001).

CONCLUSIONS

It is suggested that using complicated non-linear models like ANN could fully utilize the predictive ability of the independent variables, and then finally achieve a superior GFR estimation model.

Comparison between the Beta-2 Microglobulin-Based Equation and the CKD-EPI Equation for Estimating GFR in CKD Patients in China: ES-CKD Study

Background

Beta-2 microglobulin (B2M) and cystatin C are novel glomerular filtration markers that have a stronger association with adverse outcomes than creatinine. The B2M-based glomerular filtration rate (GFR) estimating equation was built in 2016. Several new creatinine and cystatin C equations were developed in 2019 in China. However, external validation of these new equations remains to be seen.

Methods

This is a prospective cohort study. The equations were validated in a population totaling 830 participants (median age 62 years). These equations include the B2M-based equation (built in 2016), three CKD-EPI equations (built in 2009 and 2012), three Yang-Du equations (C-CKD-EPIscr, C-CKD-EPIcys, and C-CKD-EPIscr-cys equations, all of which were Chinese-modified CKD-EPI equations developed by Yang et al. in 2019), and a Xiangya equation (a creatinine-based equation built in the Third Xiangya Hospital in 2019). The estimated GFR (eGFR) calculated separately by 8 equations (B2M GFR, CKD-EPIscr, CKD-EPIcys, CKD-EPIscr-cys, C-CKD-EPIscr, C-CKD-EPIcys, C-CKD-EPIscr-cys, and Xiangya equations) was compared with the reference GFR (rGFR) measured by the ^99mTc-DTPA renal dynamic imaging method. Participants were divided into CKD stage 1-5 specific subgroups. The primary outcomes of this study were bias, precision (interquartile range of difference, IQR), and accuracy (the proportion of eGFR within 30% of rGFR [P30] and root mean square error [RMSE]) of eGFR versus rGFR.

Results

The B2M-based equation was worse than CKD-EPI equations and Yang-Du equations in most outcomes. CKD-EPIscr and C-CKD-EPIscr equations had a larger area under the receiver operating characteristic curve (ROC^AUC). The CKD-EPIscr equation had the highest sensitivity (83.3%) and the Xiangya equation the highest specificity (89.5%) to diagnose CKD. The bias was the lowest in CKD-EPIcys and C-CKD-EPIscr-cys equations by median and mean difference (1.23 and -1.42, respectively). The Xiangya equation yielded the highest bias by both median and mean difference (8.29 and 6.52, respectively). The C-CKD-EPIscr equation was the most accurate with the highest P30 value (68.1%) and most precise with the lowest IQR (19). The Xiangya equation had the best RMSE (lowest RMSE, 0.56), and gave the best performance in the CKD stage 2 subgroup. The C-CKD-EPIscr-cys equation achieved the lowest bias in CKD stage 3-5 (p = 0.663, 0.104, and 0.130, respectively, compared with rGFR).

Conclusion

The B2M-based equation was worse than CKD-EPI and Yang-Du equations on the whole. CKD-EPIcys and C-CKD-EPIscr-cys equations had the lowest bias, whereas the Xiangya equation yielded the highest bias. The Xiangya equation gave the best performance in the CKD stage 2 subgroup, while the C-CKD-EPIscr-cys equation achieved the lowest bias in CKD stage 3-5. Further work to improve the performance of the GFR estimating equation is needed.

Role of Artificial Intelligence in Kidney Disease

Artificial intelligence (AI), as an advanced science technology, has been widely used in medical fields to promote medical development, mainly applied to early detections, disease diagnoses, and management. Owing to the huge number of patients, kidney disease remains a global health problem. Challenges remain in its diagnosis and treatment. AI could take individual conditions into account, produce suitable decisions and promise to make great strides in kidney disease management. Here, we review the current studies of AI applications in kidney disease in alerting systems, diagnostic assistance, guiding treatment and evaluating prognosis. Although the number of studies related to AI applications in kidney disease is small, the potential of AI in the management of kidney disease is well recognized by clinicians; AI will greatly enhance clinicians' capacity in their clinical practice in the future.

Glomerular Filtration Rate Estimation by a Novel Numerical Binning-Less Isotonic Statistical Bivariate Numerical Modeling Method

Statistical bivariate numerical modeling is a method to infer an empirical relationship between unpaired sets of data based on statistical distributions matching. In the present paper, a novel efficient numerical algorithm is proposed to perform bivariate numerical modeling. The algorithm is then applied to correlate glomerular filtration rate to serum creatinine concentration. Glomerular filtration rate is adopted in clinical nephrology as an indicator of kidney function and is relevant for assessing progression of renal disease. As direct measurement of glomerular filtration rate is highly impractical, there is considerable interest in developing numerical algorithms to estimate glomerular filtration rate from parameters which are easier to obtain, such as demographic and `bedside’ assays data.

Comparison of three data mining models for prediction of advanced schistosomiasis prognosis in the Hubei province

BACKGROUND

In order to better assist medical professionals, this study aimed to develop and compare the performance of three models-a multivariate logistic regression (LR) model, an artificial neural network (ANN) model, and a decision tree (DT) model-to predict the prognosis of patients with advanced schistosomiasis residing in the Hubei province.

METHODOLOGY/PRINCIPAL FINDINGS

Schistosomiasis surveillance data were collected from a previous study based on a Hubei population sample including 4136 advanced schistosomiasis cases. The predictive models use LR, ANN, and DT methods. From each of the three groups, 70% of the cases (2896 cases) were used as training data for the predictive models. The remaining 30% of the cases (1240 cases) were used as validation groups for performance comparisons between the three models. Prediction performance was evaluated using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. Univariate analysis indicated that 16 risk factors were significantly associated with a patient's outcome of prognosis. In the training group, the mean AUC was 0.8276 for LR, 0.9267 for ANN, and 0.8229 for DT. In the validation group, the mean AUC was 0.8349 for LR, 0.8318 for ANN, and 0.8148 for DT. The three models yielded similar results in terms of accuracy, sensitivity, and specificity.

CONCLUSIONS/SIGNIFICANCE

Predictive models for advanced schistosomiasis prognosis, respectively using LR, ANN and DT models were proved to be effective approaches based on our dataset. The ANN model outperformed the LR and DT models in terms of AUC.

Improving precision of glomerular filtration rate estimating model by ensemble learning

Background

Accurate assessment of kidney function is clinically important, but estimates of glomerular filtration rate (GFR) by regression are imprecise.

Methods

We hypothesized that ensemble learning could improve precision. A total of 1419 participants were enrolled, with 1002 in the development dataset and 417 in the external validation dataset. GFR was independently estimated from age, sex and serum creatinine using an artificial neural network (ANN), support vector machine (SVM), regression, and ensemble learning. GFR was measured by 99mTc-DTPA renal dynamic imaging calibrated with dual plasma sample 99mTc-DTPA GFR.

Results

Mean measured GFRs were 70.0 ml/min/1.73 m² in the developmental and 53.4 ml/min/1.73 m² in the external validation cohorts. In the external validation cohort, precision was better in the ensemble model of the ANN, SVM and regression equation (IQR = 13.5 ml/min/1.73 m²) than in the new regression model (IQR = 14.0 ml/min/1.73 m², P < 0.001). The precision of ensemble learning was the best of the three models, but the models had similar bias and accuracy. The median difference ranged from 2.3 to 3.7 ml/min/1.73 m², 30% accuracy ranged from 73.1 to 76.0%, and P was > 0.05 for all comparisons of the new regression equation and the other new models.

Conclusions

An ensemble learning model including three variables, the average ANN, SVM, and regression equation values, was more precise than the new regression model. A more complex ensemble learning strategy may further improve GFR estimates.

Electronic supplementary material

The online version of this article (10.1186/s12967-017-1337-y) contains supplementary material, which is available to authorized users.

Biomarkers of Renal Injury in Cirrhosis: Association with Acute Kidney Injury and Recovery after Liver Transplantation

BACKGROUND

To define urine or serum biomarkers in predicting renal function recovery after liver transplantation (LT).

METHODS

Adults listed for LT (February 2011-July 2014) and with modified diet for renal disease-6 (MDRD-6) <60 mL/min provided urine/blood samples at baseline and serially until LT for biomarkers in serum (pg/mL) and urine (pg/mg creatinine).

RESULTS

Of 271 LT listed patients (mean age 57 years, 63% males, median listing MELD 17.5), 1 year acute kidney injury (AKI) probability was 49%, with odds of 1.3-, 3.0-, 4.6-, and 8.5-fold times for listing MELD 16-20, 21-25, 26-30, and >30, compared to MELD <16. Thirty-seven people died over 1 year from the time of listing, with twofold increased odds with AKI. Among 67 patients with MDRD <60, only urinary epidermal growth factor was different comparing AKI (increase in serum creatinine ≥0.3 mg/dL from baseline within past 3 months) vs. no AKI (2,254 vs. 4,253, p = 0.003). Differences between acute tubular necrosis (ATN) and hepatorenal syndrome could not be ascertained for a small sample of 3 patients with ATN. Analyzing 15 of 43 receiving LT and MDRD-6 <30 prior to LT, biomarkers were not different comparing 5 patients recovering renal function (MDRD-6 >50 mL/min) at 6 months vs. 10 without recovery.

CONCLUSIONS

AKI is common among LT listed patients, with a negative impact on transplant-free survival. Serum and urine biomarkers are not associated with the recovery of renal function after LT. Multicenter studies are suggested to (a) develop strategies to reduce the development of AKI and (b) derive novel biomarkers for use in accurately predicting renal recovery after LT.

Complementarity of Clinician Judgment and Evidence Based Models in Medical Decision Making: Antecedents, Prospects, and Challenges

Early accounts of the development of modern medicine suggest that the clinical skills, scientific competence, and doctors' judgment were the main impetus for treatment decision, diagnosis, prognosis, therapy assessment, and medical progress. Yet, clinician judgment has its own critics and is sometimes harshly described as notoriously fallacious and an irrational and unfathomable black box with little transparency. With the rise of contemporary medical research, the reputation of clinician judgment has undergone significant reformation in the last century as its fallacious aspects are increasingly emphasized relative to the evidence based options. Within the last decade, however, medical forecasting literature has seen tremendous change and new understanding is emerging on best ways of sharing medical information to complement the evidence based medicine practices. This review revisits and highlights the core debate on clinical judgments and its interrelations with evidence based medicine. It outlines the key empirical results of clinician judgments relative to evidence based models and identifies its key strengths and prospects, the key limitations and conditions for the effective use of clinician judgment, and the extent to which it can be optimized and professionalized for medical use.

Development and validation of new glomerular filtration rate predicting models for Chinese patients with type 2 diabetes

Background

Previous researches has depicted that the performance of the recommended glomerular filtration rate (GFR)-estimating equations in the type 2 diabetic population is inferior to that in the non-diabetic population. We attempted to develop new GFR-predicting models for use in Chinese patients with type 2 diabetes in this study.

Methods

We enrolled 519 type 2 diabetic patients including a development data-set (n = 276), an internal validation data-set (n = 138) and an external validation data-set (n = 105) to establish new GFR-predicting models. 99mTc-DTPA-GFR revised by the dual sample method was referred to as the gold GFR standard.

Results

Based on sex, age, serum creatinine and new predictor variables [body mass index (BMI), hemoglobinA1C, and urinary albumin creatinine ratio], eight new regression models and eight artificial neural network (ANN) models were developed. In the external validation group, only ANN3 was superior in both precision and accuracy over the original CKD-EPI equation (precision, 20.5 vs. 24.2 mL/min/1.73 m², P < 0.001; 30 % accuracy, 88.6 vs. 80.6 %, P = 0.02).

Conclusions

ANN3 based on sex, age, serum creatinine and BMI is the optimal model for GFR estimation in Chinese patients with type 2 diabetes.

Electronic supplementary material

The online version of this article (doi:10.1186/s12967-015-0674-y) contains supplementary material, which is available to authorized users.

Supervised learning methods in modeling of CD4+ T cell heterogeneity

Background

Modeling of the immune system – a highly non-linear and complex system – requires practical and efficient data analytic approaches. The immune system is composed of heterogeneous cell populations and hundreds of cell types, such as neutrophils, eosinophils, macrophages, dendritic cells, T cells, and B cells. Each cell type is highly diverse and can be further differentiated into subsets with unique and overlapping functions. For example, CD4+ T cells can be differentiated into Th1, Th2, Th17, Th9, Th22, Treg, Tfh, as well as Tr1. Each subset plays different roles in the immune system. To study molecular mechanisms of cell differentiation, computational systems biology approaches can be used to represent these processes; however, the latter often requires building complex intracellular signaling models with a large number of equations to accurately represent intracellular pathways and biochemical reactions. Furthermore, studying the immune system entails integration of complex processes which occur at different time and space scales.

Methods

This study presents and compares four supervised learning methods for modeling CD4+ T cell differentiation: Artificial Neural Networks (ANN), Random Forest (RF), Support Vector Machines (SVM), and Linear Regression (LR). Application of supervised learning methods could reduce the complexity of Ordinary Differential Equations (ODEs)-based intracellular models by only focusing on the input and output cytokine concentrations. In addition, this modeling framework can be efficiently integrated into multiscale models.

Results

Our results demonstrate that ANN and RF outperform the other two methods. Furthermore, ANN and RF have comparable performance when applied to in silico data with and without added noise. The trained models were also able to reproduce dynamic behavior when applied to experimental data; in four out of five cases, model predictions based on ANN and RF correctly predicted the outcome of the system. Finally, the running time of different methods was compared, which confirms that ANN is considerably faster than RF.

Conclusions

Using machine learning as opposed to ODE-based method reduces the computational complexity of the system and allows one to gain a deeper understanding of the complex interplay between the different related entities.

Seasonal variation in onset and relapse of IBD and a model to predict the frequency of onset, relapse, and severity of IBD based on artificial neural network

BACKGROUND

Previous research has yielded conflicting data as to whether the natural history of inflammatory bowel disease follows a seasonal pattern. The purpose of this study was (1) to determine whether the frequency of onset and relapse of inflammatory bowel disease follows a seasonal pattern and (2) to establish a model to predict the frequency of onset, relapse, and severity of inflammatory bowel disease (IBD) with meteorological data based on artificial neural network (ANN).

METHOD

Patients with diagnosis of ulcerative colitis (UC) or Crohn's disease (CD) between 2003 and 2011 were investigated according to the occurrence of onset and flares of symptoms. The expected onset or relapse was calculated on a monthly basis over the study period. For artificial neural network (ANN), patients from 2003 to 2010 were assigned as training cohort and patients in 2011 were assigned as validation cohort. Mean square error (MSE) and mean absolute percentage error (MAPE) were used to evaluate the predictive accuracy.

RESULTS

We found no seasonal pattern of onset (P = 0.248) and relapse (P = 0.394) among UC patients. But, the onset (P = 0.015) and relapse (P = 0.004) of CD were associated with seasonal pattern, with a peak in July and August. ANN had average accuracy to predict the frequency of onset (MSE = 0.076, MAPE = 37.58%) and severity of IBD (MSE = 0.065, MAPE = 42.15%) but high accuracy in predicting the frequency of relapse of IBD (MSE = 0.009, MAPE = 17.1%).

CONCLUSION

The frequency of onset and relapse in IBD showed seasonality only in CD, with a peak in July and August, but not in UC. ANN may have its value in predicting the frequency of relapse among patients with IBD.