Neural Network Applications in Medical Research

Neural networks provide significant benefits in medical research. They are actively being used for such applications as locating previously undetected patterns in mountains of research data, controlling medical devices based on biofeedback, and detecting characteristics in medical imagery.

Below we have listed some of the common applications of neural networks in medical research. If you are currently using neural networks in your medical application, we would love to hear about it.

NeuroDimension has also used its leading edge neural network technology to develop numerous medical applications with a variety of companies. If you need neural network consulting for your medical application, please contact NeuroDimension.

Sample Applications

Locate common characteristics in large amounts of data (divide research populations).
Better forecast results based on existing data (recovery time, changes to device settings).
Predict the progression of medical data over time (cell growth, disease dispersion).
Identify specific characteristics in medical imagery (ultrasound/x-ray feature detection).
Group medical data based on key characteristics (demographics, pre-existing conditions).

Locate common characteristics in large amounts of data.

Locating common characteristics in large amounts of data is a type of classification problem. Neural networks can be used to solve classification problems, typically through Multi-Layer Perceptron (MLP) and Support Vector Machines (SVM) type networks.

Examples of classification applications in medicine include dividing research populations or data into groups for further study. For example, data from studies of body movement could be classified into different patterns to aid with physical therapy.

Sample Study: Classification of Hand Movements

By recording EEGs during sequences of periodic left or right hand movements; patterns can be found, classified in real time, and used to move a cursor on a monitor left or right. EEG classification is necessary when the EEG is used as an input signal to a brain computer interface (BCI). Such a BCI can be helpful for handicapped people. In this study 4 out of 6 subjects showed a classification accuracy of 89-100%. Electroencephalography and Clinical Neurophysiology (1996) G Pfurtscheller, J Kalcher, C Neuper, D Flotzinger … Locate this paper on Google Scholar!

Sample Video: Breast Cancer Detection

Using NeuroSolutions, this project can be used to illustrate how easy it is to get started with a real world application in a matter of minutes and get great results out of the box! This breast cancer databases was obtained from the University of Wisconsin Hospitals, Madison from Dr. William H. Wolberg. For more information on the data set, please refer to the UCI Machine Learning Data Repository. This text will be replaced

Sample Project: Breast Cancer Detection

The following dataset was featured in the video demonstration above. Use it with a free evaluation of NeuroSolutions. Download the dataset: cancer.zip Use with NeuroSolutions: Free Evaluation

Our NeuroSolutions product is an excellent resource for classification applications. For an interactive example of classification in NeuroSolutions for Excel, download the free evaluation version and view the demo called “Testing Classifiers” in the Help menu.

Better forecast results based on existing data.

Forecasting results based on existing data is a type of function approximation problem. Neural networks can be used to solve function approximation problems, typically through Multi-Layer Perceptron (MLP), Radial Basis Function (RBF) and CANFIS (Co-Active Neuro-Fuzzy Inference System) type networks.

Examples of function approximation applications in medicine include the prediction of patient recovery and automated changes to device settings. For example, data from studies of potential recovery level of patients can provide realistic estimates to patients while helping facilities cut costs by better allocating resources.

Sample Study: Functional recovery of stroke survivors

One study was able to use neural networks to predict with 88% accuracy the Functional Independence Measure (FIM) score among stroke survivors with moderate disabilities on discharge based on a small set of clinical variables and the admission FIM score. Neural network modeling accurately predicts the functional outcome of stroke survivors with moderate disabilities (1997) WJ Oczkowski, S Barreca - Arch Phys Med Rehabil Locate this paper on Google Scholar!

Sample Project: Body Fat.

Using NeuroSolutions, this project can be used to illustrate multiple regression techniques. Accurate measurement of body fat is inconvenient/costly and it is desirable to have easy methods of estimating body fat that are not inconvenient/costly. Download the dataset: Bodyfat.zip, 520KB Use with NeuroSolutions: Free Evaluation

Our NeuroSolutions product is an excellent resource for function approximation applications. For an interactive example of function approximation in NeuroSolutions, download the free evaluation version of the software and view the demo called “Multi-Layer Perceptron, Basic” in the Help menu.

Predict the progression of medical data over time.

Predicting the progression of medical data over time is a type of time-series prediction problem. Neural networks can be used to solve time-series problems, typically through Time-Lagged Recurrent (TLRN) type network.

Examples of time-series predictions in medicine include the prediction of cell growth and disease dispersion. For example, data from studies of muscle stimulation patterns of arm movements can be used to control mouse movements on a computer screen.

Sample Study: Control of Arm Movements Application

Using an artificial neural network (ANN) controller to learn and store optimal patterns of muscle stimulation for a range of single joint movements, one study found a new approach to the control of point to point, single joint arm movements. Stimulation patterns that minimize muscle activation or muscular effort are obtained from an optimal control strategy. Neural network topologies considered in this study are feed forward, recurrent feedback, and time delay. IEEE Transactions on Rehabilitation Engineering (1994) N Lan, HQ Feng, PE Crago Locate this paper on Google Scholar!

Our NeuroSolutions product is an excellent resource for time-series prediction applications. For an interactive example of time-series prediction in NeuroSolutions, download the free evaluation version of the software and view the demo called “Time Lagged Recurrent Network” in the Help menu.

Identify specific characteristics in medical imagery.

Identifying specific characteristics in medical imagery is a type of image processing problem. Neural networks can be used to solve image processing problems, typically through Principal Component Analysis (PCA) type network.

Examples of image processing in medicine include the detection of characteristics in ultrasound and x-ray features. For example, image data from studies of mammograms can be used for the detection of breast cancer.

Sample Study: Mammography

One study found that neural networks provide a useful tool to aid radiologists in the mammography decision making task. With clinical cases, the performance of a neural network in features extracted of lesions from mammograms by radiologist was found to be higher in distinguishing between benign and malignant lesions than average performance of radiologist alone, without the aid of a neural network. Artificial neural networks in mammography: application to decision making in the diagnosis of breast cancer (1993) Y Wu, ML Giger, K Doi, CJ Vyborny, RA Schmidt, CE… Locate this paper on Google Scholar!

Our NeuroSolutions product is an excellent resource for image processing applications. For an interactive example of image processing in NeuroSolutions, download the free evaluation version of the software and view the demo called “Linear Associator” in the Help menu.

Group medical data based on key characteristics.

Grouping of medical data based on key characteristics is a type of clustering problem. Neural networks can be used to solve clustering problems, typically through Self-Organizing Map (SOM) type network.

Examples of clustering in medicine include the detection of key characteristics in demographics or pre-existing conditions. For example, data from studies combined with sensitivity analysis can reverse engineer a biologically plausible relationship from real world data.

Sample Study: Temporal Gene Expression Data

Temporal gene data can be used to create gene networks where they would be used to represent regulatory interactions between genes over time. The reverse engineering of gene networks & extraction of regulatory relationships between genes in temporal gene expressions data is a major obstacle in systems biology. Neural networks and temporal gene expression data - A Krishna, A Narayanan, EC Keedwell Locate this paper on Google Scholar!

Our NeuroSolutions product is an excellent resource for clustering applications. For an interactive example of clustering in NeuroSolutions, download the free evaluation version of the software and view the demo called “Unsupervised Learning” in the Help menu.