UroComp Home
Go to UroComp home

Hypogonadism Detection Neural Network

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

Predicting hypogonadism in men based upon age, presence of erectile dysfunction, and depression. A Kshirsagar, A Seftel, L Ross, M Mohamed, and C Niederberger, International Journal of Impotence Research (www.ijir.com) August 4, 2005.

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 Hypogonadism

Hypogonadism, a persistent low serum testosterone level, is a disorder that can occur at any age but has an increased incidence with aging. A persistent low testosterone level can lead to a decreased libido, musculoskeletal decline, increased body fat and has been associated with erectile dysfunction (ED), depression, as well as cognitive impairment. Identification of this disorder is challenging as all the manifestations of hypogonadism may not be present in any one individual or be present with varied levels of severity. Additionally, these signs and symptoms often can be attributed to the aging process or other disease states.

Age, the presence of ED and depression are three risk factors that can be easily ascertained in the outpatient setting. ED can be assessed by the Sexual Health Inventory for Men (SHIM) score. This validated subset of the IIEF (International Index of Erectile Dysfunction) is a 5-item questionnaire whose aggregate score ranges from 0-25: higher scores indicate better erectile function. Depression can also be assessed with validated questionnaires. The Center for Epidemiologic Studies Depression Scale (CES-D) is a widely used depression inventory that consists of 20 items scored on a 4 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 hypogonadism in certain clinical settings involving variable combinations of related factors: age, the presence of ED and depression. This task may be accomplished by a neural network.

Return to top of document


Neural Network Programming and Training

Clinical data was collected from 218 men which served as the dataset for the analysis. This dataset was modeled using, “neUROn2++”, a set of C++ programs designed to implement neurological 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 included age, CES-D score (the depression metric), and SHIM score (the metric for ED) which were encoded as the input nodes to the neural network. The serum total testosterone level was determined to be the single discrete output node. Hypogonadism was defined as a total serum testosterone level < 300ng/dl and for outcomes modeling expressed as a binary parameter.

The dataset was randomly divided into a training set of 148 subjects and a test set of 70 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 the optimal network architecture which maintained an acceptable goodness-of-fit without overlearning. The training method was canonical off-line backpropagation with the weight decay term lambda chosen to be 5e-05. 1 The network was determined to be trained to completion when the error was observed to be oscillating at a local error minimum.

Wilk’s Generalized Likelihood Ration Test (GLRT) was applied to determine which input features were significant to the model’s outcome in a reverse regression analysis.1 The presence of depression (CES-D score) was found to be the most significant risk factor (p < 0.002). The next most significant risk factor (p < 0.007) was determined to be the presence of ED (SHIM score). A patient’s age had the least influence on the network but still had statistical value (p < 0.02). The dataset was also modeled using logistic regression, linear discriminant function analysis, and quadratic discriminant function analysis to compare traditional linear statistical modeling tools with the nonlinear computational neural network.

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
0.725
0.725
Logistic Regression
0.619
0.609
Quadratic Discriminant Function Analysis
0.580
0.075
Linear Discriminant Function Analysis
0.574
0.215

  1. A 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.

  2. Receiver Operating Characteristic Curve area. As numbers approach 1.0, the accuracy of the statistical method improves: a ROC value of 1.0 would indicate a sensitivity of 1.0 and specificity 1.0. ROC 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 TPredict.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