Artificial Neural Network Applications in Medicine

Doctors have an extensive amount of medical data at their disposal. However, that data is too large for humans to quickly organize, analyze, and identify patterns. Today, doctors are beginning to use artificial neural networks (ANN) to help catch patterns to aid in the diagnostic process. This research paper will discuss the beneficial applications of artificial neural networks currently and down the road.

Today Doctors are trying to implement fast, accurate, and cost-effective ways of diagnosing their patients. Over the past decade, with advancements in computing power and artificial intelligence algorithms, doctors have begun to turn to artificial neural networks to aid in the identification of cancers, cardiovascular disease, and diabetes. Some artificial neural networks have had an accuracy of over 90% for the diagnosis of numerous types of disease.

Artificial neural networks are a mathematical representation of the human brain. The neural network is formed by nodes, which are organized in layers similar to how neurons are in the human brain. A traditional ANN has a single input layer and a single output layer. As the names imply, the input layer is where the user enters data for the network to process, while the output layer is the end result(s) of the network. The input and output layers are the first and last layers, respectively, on an ANN. Between the input and output layers is at least one hidden layer, as seen in figure 1.

Every node in the input layer has a weighted link to every node in the first hidden layer. The same thing is true for every node in every hidden layer. There can be any number of nodes per layer and any number of total hidden layers. It depends on the complexity of what is being studied. The hidden layers will process data given from the previous layer. This is done mathematically. The outputs of the network are processed in the last layer, which doubles as the output layer. Figure 1 above depicts a neural network with two hidden layers. The figures show that each subsequent layer contains fewer neurons. The number of possible outputs is also small when compared to the number of inputs from the input layer.

As data goes through a neural network, the network learns, and then it can output its findings to the user. The neural network is given a training database made up of real-life examples of medical data and diagnoses associated with the specific condition that the ANN aims to diagnose. As the neural network gives more information about previous cases, it becomes more accurate with its own predictions. The ANN is able to compare thousands of known cases with the case they are looking to diagnose. Figure 2 below is an abstract example of a training database.

Each row consists of elements that fall under one of two categories: medical data and diagnosis. Patient code refers to the fact that the data and diagnosis in the given row correspond to that of an existing patient case. The medical data consists of symptoms, medical tests, pre-existing medical conditions, etc. The same medical data is being compared for all patients, but not all patient data is identical. Diagnosis is either positive or negative in this example, but it can also be uncertain. The ANN will then use the medical data and the diagnosis to find patterns and trends between patients who had a positive diagnosis and see how they differ from patients with a negative diagnosis.

The user of an artificial neural network sees none of the algorithms being conducted by the network. To the user, in this case, a doctor, the neural network seems like a “black box.” The user will input a series of data, receive a series of outputs, and else. The average doctor does not readily know the math conducted by the ANN, nor does this benefit them. A “black box” is a term given to a device that is viewed only in terms of inputs and outputs. This makes ANN very user-friendly to doctors. A doctor’s expertise is not in understanding how an ANN works or how to build one. That is up to the programmer and engineers designing it.

When doctors use artificial neural networks to aid in the diagnostic process, they enter symptoms, bio test analysis (blood, urine tests, etc.), and other personal characteristics (age, gender, substance use). After all, this is entered, the neural network goes into action, comparing it with trends and patterns of known diseases. The network then outputs the result of the diagnosis. The result is usually positive, negative, or uncertain. This process is much faster, more efficient, and more accurate than a conventional human diagnosis. ANNs used for helping diagnose breast cancer have taken less than 10 minutes to train, simulate and diagnose if a patient may have breast cancer. Often, doctors need to consult other doctors to figure out what is going on. If doctors give a false diagnosis, they sometimes do not know it was incorrect until they have started treating the patient for what they initially thought.

A group of researchers from Pharos University, Alexandria University, and the Alexandria Institute of Engineering and Technology conducted research detailing the use of neural networks in the detection of breast cancer tumors. The group was using ultra-wideband microwave (UWB) signals in conjunction with an artificial network to accurately detect tumors in the model breast. The scattering of the UWB signals off breast tissue was collected and stored in a database to be used for the training of the ANN.

The ANN used in the experiment was an optimized feedforward propagation neural network designed and written entirely in MATLAB and consisted of only one hidden layer. The researchers used feedforward propagation for its efficiency when it came to their small sets of data. The ANN in this experiment was able to output the location of the tumor in the model breast. This is very exciting because the neural network allowed for the classification and location detection of the tumor without the need for additional medical screenings or expert input.

The applications and benefits of artificial neural networks in the medical field are already evident. Easy to use and train, they allow for the easy organization of patient data and the accurate detection of ailments. Compared to medical experts, they are better for the rapid recognition of patterns in medical data. This allows for people to be diagnosed earlier and faster. It is also useful for use in conjunction with doctors or other medical tools. I would like to acknowledge Sam Klomp for his help during the process of this research paper. Sam played a big part in the research, proposal, and rough draft of this assignment.

Cite this page