UroComp Home
Go to UroComp home

Erectile Dysfunction Detection Neural Network

Portions of this document were adapted in part from our paper:

Wald M, Seftel AD, Ross LS, Mohamed MAB, Niederberger CS: Computational Models for Detection of Erectile Dysfunction: Journal of Urology, Volume 173, January 2005, pp 167-70

The reader is strongly encouraged to read this paper prior to using the network, as it contains a more specific description of the model.


Contents


About Erectile Dysfunction Detection

Erectile Dysfunction (ED) has been identified as one of the four major, non-cancer disease states that adversely effect men over the age of 50. The multifactorial nature suggested for ED may be explained by its cause-and-effect relationships with other comorbidities, including benign prostatic hyperplasia, cardiovascular disease and depression, as well as by other clinical associations. ED is considered as age-related by some authors, and has been shown to correlate with depressive symptoms. In addition, hormonal abnormalities are estimated to exist in about 10% of patients with ED. Serum testosterone levels decline slowly with normal male aging. However, while testosterone impacts libido, the role of low testosterone levels in ED is not as clear.

Erectile dysfunction can be assessed by the Sexual Health Inventory for Men (SHIM) score. This validated subset of the IIEF (International Index of Erectile Function) is a 5-item questionnaire, whose aggregate score ranges from 0 to 25, where higher scores indicate better erectile function. The CES-D is a widely used depression inventory that consists of 20 items scored on a four-point scale. A score of 16 and above is considered to indicate the existence of a depressive disorder.

Clinicians may benefit from a tool which predicts erectile dysfunction in certain clinical settings, involving variable combinations of related factors. This task may be accomplished by a neural network. Furthermore, statistical analysis of the neural network may determine the significance of the evaluated clinical factors to the model's outcome.

Return to top of document


Neural Network Programming and Training

Clinical data collected from 140 men seeking treatment for erectile dysfunction was used to construct the dataset for the study. This dataset was modeled using "neUROn2++", a suite of C++ programs we designed to implement neural computational and statistical algorithms, which are cross compiled using Microsoft Visual C++ version 6 and GNU C++ (Cygwin port version 2.95).

The 3 clinical features evaluated include age, total testosterone level and the depression metric CES-D, which were encoded as the input nodes to the neural network. The output node represented the Sexual Health Inventory for men (SHIM) score, which was encoded as a binary variable (presence or absence of severe erectile dysfunction). For outcomes modeling, a score of 10 and below was assigned 1 and a score higher than 10 was assigned 0

The dataset was randomly divided into a training set of 105 subjects, and a test set of 35 subjects. The test set was excluded from training, and only used for cross-validation (n1/n2 method). A 1-hidden node layer with 4 nodes was determined to represent an optimal network architecture which maintained acceptable goodness-of-fit without overlearning. The training method was canonical off-line backpropagation with weight decay, with the weight decay term lambda chosen to be 5e-05.1 The network to be trained to completion when the error was observed to be oscillating at a local error minimum.

Wilk's Generalized Likelihood Ratio Test (GLRT) was applied to determine which input features were significant to the model's outcome in a forward and reverse regression analysis.1 Age was found to be most significant (p< 0.001), followed by CES-D score (p < 0.03), followed by testosterone (p > 0.6). The dataset was also modeled using logistic regression and linear and quadratic discriminant function analysis (LDFA and QDFA) to compare the nonlinear computational method of neural computation with traditional linear statistical modeling tools.

1A description of this method of training, including weight decay and feature extraction using Wilk's GLRT, may be found in Golden RM, Mathematical methods for neural network analysis and design, Cambridge, MA: MIT Press, 1996.

Return to top of document


Accuracy of the Neural Network Compared to Linear Methods

Method
Training Set ROC Area2 Test Set ROC Area2
Neural Network
0.798
0.702
Logistic Regression
0.661
0.618
Quadratic Discriminant Function Analysis
0.679
0.676
Linear Discriminant Function Analysis
0.629
0.645

2Receiver Operating Characteristic Curve area. Numbers approach 1.0 as accuracy improves (a value of 1.0 would indicate sensitivity and specificity both 1.0). Areas were computed using the statistical method described by Wickens: Wickens TD, Elementary signal detection theory, New York: Oxford University Press, 2002.

Return to top of document


Click here to download EDPredict.prc, a PalmOS application for the PalmPilot or Handspring Visor device. (If you are using Netscape Navigator, and have trouble with this link, try clicking on it with the right mouse button, and choosing "Save Link As..." from the pop-up menu.)

  

Return to top of document