Artificial neural networks for prostate carcinoma risk assessment. An overview

Artificial neural network-based models used for predicting 28- and 90-day mortality of patients with hepatitis B-associated acute-on-chronic liver failure

Background

This study aimed to develop prognostic models for predicting 28- and 90-day mortality rates of hepatitis B virus (HBV)-associated acute-on-chronic liver failure (HBV-ACLF) through artificial neural network (ANN) systems.

Methods

Six hundred and eight-four cases of consecutive HBV-ACLF patients were retrospectively reviewed. Four hundred and twenty-three cases were used for training and constructing ANN models, and the remaining 261 cases were for validating the established models. Predictors associated with mortality were determined by univariate analysis and were then included in ANN models for predicting prognosis of mortality. The receiver operating characteristic curve analysis was used to evaluate the predictive performance of the ANN models in comparison with various current prognostic models.

Results

Variables with statistically significant difference or important clinical characteristics were input in the ANN training process, and eight independent risk factors, including age, hepatic encephalopathy, serum sodium, prothrombin activity, γ-glutamyltransferase, hepatitis B e antigen, alkaline phosphatase and total bilirubin, were eventually used to establish ANN models. For 28-day mortality in the training cohort, the model’s predictive accuracy (AUR 0.948, 95% CI 0.925–0.970) was significantly higher than that of the Model for End-stage Liver Disease (MELD), MELD-sodium (MELD-Na), Chronic Liver Failure-ACLF (CLIF-ACLF), and Child-Turcotte-Pugh (CTP) (all p < 0.001). In the validation cohorts the predictive accuracy of ANN model (AUR 0.748, 95% CI: 0.673–0.822) was significantly higher than that of MELD (p = 0.0099) and insignificantly higher than that of MELD-Na, CTP and CLIF-ACLF (p > 0.05). For 90-day mortality in the training cohort, the model’s predictive accuracy (AUR 0.913, 95% CI 0.887–0.938) was significantly higher than that of MELD, MELD-Na, CTP and CLIF-ACLF (all p < 0.001). In the validation cohorts, the prediction accuracy of the ANN model (AUR 0.754, 95% CI: 0.697–0.812 was significantly higher than that of MELD (p = 0.019) and insignificantly higher than MELD-Na, CTP and CLIF-ACLF (p > 0.05).

Conclusions

The established ANN models can more accurately predict short-term mortality risk in patients with HBV- ACLF.

The main content has been postered as an abstract at the AASLD Hepatology Conference (10.1002/hep.30257).

Prediction model of artificial neural network for the risk of hyperuricemia incorporating dietary risk factors in a Chinese adult study

Background

Risk of hyperuricemia (HU) has been shown to be strongly associated with dietary factors. However, there is scarce evidence on prediction models incorporating dietary factors to estimate the risk of HU.

Objective

The aim of this study was to develop a prediction model to predict the risk of HU in Chinese adults based on dietary information.

Design

Our study was based on a cross-sectional survey, which recruited 1,488 community residents aged 18 to 60 years in Beijing from October 2010 to January 2011. The eligible participants were randomly divided into a training set (n₁ = 992) and a validation set (n₂ = 496) in the ratio of 2:1. We developed the prediction model in three stages. We first used a logistic regression model (LRM) based on the training set to select a set of dietary risk factors which were related to the risk of HU. Artificial neural network (ANN) was then used to construct the prediction model using the training set. Finally, we used receiver operating characteristic (ROC) curve analysis to assess the accuracy of the prediction model using training and validation sets.

Results

In the training set, the mean age of participants with and without HU was 39.3 (standard deviation [SD]: 9.65) and 38.2 (SD: 9.38) years, respectively. Patients with HU consisted of 101 males (77.7%) and 29 females (22.3%). The LRM found that food frequency (vegetables [odds ratio (OR) = 0.73], meat [0.72], eggs [0.80], plant oil [0.78], tea [0.51], eating habits (breakfast [OR = 1.28]), and the salty cooking style (OR = 1.33) were associated with risk of HU. In the ANN analysis, we selected a three-layer back propagation neural network (BPNN) model with 14, 3, and 1 neuron in the input, hidden, and output layers, respectively, as the best prediction model. The areas under the ROC of the training and validation sets were 0.827 and 0.814, respectively. HU would occur when the incidence probability is greater than 0.128. The indicators of accuracy, sensitivity, specificity, and Yuden Index suggested that the ANN model in our study is successful and valuable.

Conclusions

This study suggests that the ANN model could be used to predict the risk of HU in Chinese adults. Further prospective studies are needed to improve the accuracy and to generalize the use of model.

A model to predict 3-month mortality risk of acute-on-chronic hepatitis B liver failure using artificial neural network

Model for end-stage liver disease (MELD) scoring was initiated using traditional statistical technique by assuming a linear relationship between clinical features, but most phenomena in a clinical situation are not linearly related. The aim of this study was to predict 3-month mortality risk of acute-on-chronic hepatitis B liver failure (ACHBLF) on an individual patient level using an artificial neural network (ANN) system. The ANN model was built using data from 402 consecutive patients with ACHBLF. It was trained to predict 3-month mortality by the data of 280 patients and validated by the remaining 122 patients. The area under the curve of receiver operating characteristic (AUROC) was calculated for ANN and MELD-based scoring systems. The following variables age (P < 0.001), prothrombin activity (P < 0.001), serum sodium (P < 0.001), total bilirubin (P = 0.015), hepatitis B e antigen positivity rate (P < 0.001) and haemoglobin (P < 0.001) were significantly related to the prognosis of ACHBLF and were selected to build the ANN. The ANN performed significantly better than MELD-based scoring systems both in the training cohort (AUROC = 0.869 vs 0.667, 0.591, 0.643, 0.571 and 0.577; P < 0.001, respectively) and in the validation cohort (AUROC = 0.765 vs 0.599, 0.563, 0.601, 0.521 and 0.540; P ≤ 0.006, respectively). Thus, the ANN model was shown to be more accurate in predicting 3-month mortality of ACHBLF than MELD-based scoring systems.

Artificial neural network modeling for deciding if extractions are necessary prior to orthodontic treatment

OBJECTIVE

To construct a decision-making expert system (ES) for the orthodontic treatment of patients between 11 and 15 years old to determine whether extraction is needed by using artificial neural networks (ANN). Specifically, we will uncover the factors that affect this decision-making process.

METHODS

A total of 200 subjects were chosen; among them, 120 were accepted for extraction treatments, and 80 were chosen for nonextraction treatments. For each case, 23 indices were selected. A 23-13-1 Back Propagation (BP) ANN model was constructed, and the data for 180 patients were aggregated to constitute the training set. Data for the other 20 patients were used as the testing set.

RESULTS

When data from the 180 patients that had been trained were tested, the result was 100%, as expected. The untrained data from 20 patients in the testing set were 80% correct (ie, 16 cases were forecasted successfully). In the meantime, the relative contributions of the 23 input indices to the final output index (extraction/nonextraction) were calculated. "Anterior teeth uncovered by incompetent lips" and "IMPA (L1-MP)" were the two indices that gave the biggest contributions sequentially; the index of FMA (FH-MP) gave the smallest contribution.

CONCLUSIONS

(1) The constructed artificial neural network in this study was effective, with 80% accuracy, in determining whether extraction or nonextraction treatment was best for malocclusion patients between 11 and 15 years old; (2) when the clinician is predicting whether an orthodontic treatment requires extraction, the indices "anterior teeth uncovered by incompetent lips" and "IMPA (L1-MP)" should be taken into consideration first.

Outcome prediction for prostate cancer detection rate with artificial neural network (ANN) in daily routine

BACKGROUND

We evaluated the use of the artificial neural network (ANN) program "ProstataClass" of the Department of Urology and the Institute of Medical Informatics at the Charité-Universitätsmedizin Berlin in daily routine to increase prostate cancer (CaP) detection rate and to reduce unnecessary biopsies.

MATERIALS AND METHODS

From May 2005 to April 2007, a total of 204 patients were included in the study. The Beckman Access PSA assay was used, and pretreatment prostate specific antigen (PSA) was measured prior to digital rectal examination (DRE) and 12 core systematic transrectal ultrasound (TRUS) guided biopsies. The individual ANN predictions were generated with the use of the ANN application for the Beckman Access PSA and free PSA assays, which relies on age, PSA, percent free prostate specific antigen (%fPSA), prostate volume, and DRE. Diagnostic validity of total prostate specific antigen (tPSA), %fPSA, and the ANN was evaluated by ROC curve analysis.

RESULTS

PSA and %fPSA ranged from 4.01 to 9.91 ng/ml (median: 6.65) and 5% to 48% (median: 15%), respectively. Of all men, 46 (22.5%) demonstrated suspicious DRE findings. Total prostate volume ranged from 7.1 to 119.2 cc (median: 35). Overall, 71 (34.8%) CaP were detected. Of men with suspicious DRE, 28 (60.9%) had CaP on initial biopsy. The ANN was 78% accurate in the original report. The AUC of ROC curve analysis was 0.51 for PSA, 0.66 for %PSA, and 0.72 for the ANN-Output, respectively.

CONCLUSIONS

Our results in this independent cohort show that ANN is a very helpful parameter in daily routine to increase the CaP detection rate and reduce unnecessary biopsies.

Prognostic accuracy of an artificial neural network in patients undergoing radical cystectomy for bladder cancer: a comparison with logistic regression analysis

OBJECTIVE

To compare the prognostic performance of an artificial neural network (ANN) with that of standard logistic regression (LR), in patients undergoing radical cystectomy for bladder cancer.

PATIENTS AND METHODS

From February 1982 to February 1994, 369 evaluable patients with non-metastatic bladder cancer had pelvic lymph node dissection and radical cystectomy for either stage Ta-T1 (any grade) tumour not responding to intravesical therapy, with or with no carcinoma in situ, or stage T2-T4 tumour. LR analysis based on 12 variables was used to identify predictors of overall 5-year survival, and the ANN model was developed to predict the same outcome. The LR analysis, based on statistically significant predictors, and the ANN model were the compared for their accuracy in predicting survival.

RESULTS

The median age of the patients was 63 years, and overall 201 of them died. The tumour stage and nodal involvement (both P<0.001) were the only statistically independent predictors of overall 5-year survival on LR analysis. Based on these variables, LR had a sensitivity and specificity for predicting survival of 68.4% and 82.8%, respectively; corresponding values for the ANN were 62.7% and 86.1%. For LR and ANN, the positive predictive values were 78.6% and 76.2%, and the negative predictive values were 73.9% and 76.5%, respectively. The index of diagnostic accuracy was 75.9% for LR and 76.4% for ANN.

CONCLUSIONS

The ANN accurately predicted the survival of patients undergoing radical cystectomy for bladder cancer and had a prognostic performance comparable with that of LR. As ANNs are based on easy-to-use software that can identify nonlinear interactions between variables, they might become the preferred tool for predicting outcome.

Prognostic models in patients with non-small-cell lung cancer using artificial neural networks in comparison with logistic regression

It is difficult to precisely predict the outcome of each individual patient with non-small-cell lung cancer (NSCLC) by using conventional statistical methods and ordinary clinico-pathological variables. We applied artificial neural networks (ANN) for this purpose. We constructed a prognostic model for 125 NSCLC patients with 17 potential input variables, including 12 clinico-pathological variables (age, sex, smoking index, tumor size, p factor, pT, pN, stage, histology) and 5 immunohistochemical variables (p27 percentage, p27 intensity, p53, cyclin D1, retinoblastoma (RB)), by using the parameter-increasing method (PIM). Using the resultant ANN model, prediction was possible in 104 of 125 patients (83%, judgment ratio (JR)) and accuracy for prediction of survival at 5 years was 87%. On the other hand, JR and survival prediction accuracy in the logistic regression (LR) model were 37% and 78%, respectively. In addition, ANN outperformed LR for prediction of survival at 1 or 3 years. In these cases, PIM selected p27 intensity and cyclin D1 for the 3-year survival model and p53 for the 1-year survival model in addition to clinico-pathological variables. Finally, even in an independent validation data set of 48 patients, who underwent surgery 10 years later, the present ANN model could predict outcome of patients at 5 years with the JR and accuracy of 81% and 77%, respectively. This study demonstrates that ANN is a potentially more useful tool than conventional statistical methods for predicting survival of patients with NSCLC and that inclusion of relevant molecular markers as input variables enhances its predictive ability.

Multicenter Evaluation of an Artificial Neural Network to Increase the Prostate Cancer Detection Rate and Reduce Unnecessary Biopsies

Background: The percentage of free prostate-specific antigen (%fPSA) has been shown to improve specificity for the diagnosis of prostate cancer (PCa) over total PSA (tPSA). A multicenter study was performed to evaluate the diagnostic value of a %fPSA-based artificial neural network (ANN) in men with tPSA concentrations between 2 and 20 μg/L for detecting patients with increased risk of a positive prostate biopsy for cancer.

Methods: We enrolled 1188 men from six different hospitals with PCa or benign prostates between 1996 and 2001. We used a newly developed ANN with input data of tPSA, %fPSA, patient age, prostate volume, and digital rectal examination (DRE) status to calculate the risk for the presence of PCa within different tPSA ranges (2–4, 4.1–10, 2–10, 10.1–20, and 2–20 μg/L) at the 90% and 95% specificity or sensitivity cutoffs, depending on the tPSA concentration. ROC analysis and cutoff calculations were used to estimate the diagnostic improvement of the ANN compared with %fPSA alone.

Results: In the low tPSA range (2–4 μg/L), the ANN detected 72% and 65% of cancers at specificities of 90% or 95%, respectively. At 4–10 μg/L tPSA, the ANN detected 90% and 95% of cancers with specificities of 62% and 41%, respectively. Use of the ANN with 2–10 μg/L tPSA enhanced the specificity of %fPSA by 20–22%, thus reducing the number of unnecessary biopsies.

Conclusions: Enhanced accuracy of PCa detection over that obtained using %fPSA alone can be achieved with a %fPSA-based ANN that also includes clinical information from DRE and prostate volume measurements.