Artificial Neural Networks in Healthcare
Artificial Neural Networks in Healthcare – with illustration in cancer screening and diagnosis
Introduction:
Pattern recognition algorithms analyse large sets of data and use machine learning techniques to identify patterns and trends in data. These algorithms are used in a variety of fields such as facial expressions recognition, speech recognition, classification, healthcare, GIS, remote sensing, image analysis, computer vision, speech recognition, natural language processing, and data mining etc. Based on the patterns that have been identified these can be used to classify new data based on its similarity to existing patterns. There are different types of pattern recognition algorithms, including supervised learning algorithms, unsupervised learning algorithms, and semi-supervised learning algorithms. Supervised learning algorithms rely on labelled data to train the algorithm to recognize patterns, while unsupervised learning algorithms use unlabelled data to identify patterns. Semi-supervised learning algorithms use a combination of labelled and unlabelled data to improve the accuracy of pattern recognition [1].
Artificial neural networks (ANNs) are computer programs that emulate the processes of the human brain and are used to analyse and classify complex data sets by identifying patterns and relationships between variables. ANNs are composed of interconnected nodes (neurons) that process and transmit information. Each node has an input, which it receives from other neurons or the environment, and an output, which it uses to communicate with other neurons or the environment. Nodes are arranged in layers, with each layer performing a specific type of processing on the input. The input layer is the first layer of neurons in the network, and it receives data from an external source such as an image or a set of numerical values. The output of the input layer is then passed to one or more hidden layers, which perform complex processing on the input data using mathematical functions like matrix multiplications and nonlinear activation functions. Finally, the output of the hidden layers is passed to the output layer, which produces the final output of the network. During training, an ANN is presented with a set of input data, along with the corresponding desired output. The network adjusts the weights of the connections between the neurons in order to minimize the difference between the predicted output and the actual output. This process is called backpropagation, and it allows the network to learn from the data and improve its accuracy over time. One of the strengths of ANNs is their ability to learn and generalize from data, allowing them to make predictions and classify new data that they have not seen before. They are used in a wide range of applications, including computer vision, natural language processing, and speech recognition. However, ANNs can be computationally intensive and require large amounts of training data, which can make them difficult to implement in some applications [2].
Nodes or Neurons can be organized in any topological manner (e.g. one- or two-dimensional layers, three-dimensional blocks or more-dimensional structures), depending on the quality and amount of input data. Once an ANN has been trained with suitable data to find the hidden rules governing a certain phenomenon, it is then able to correctly generalize data it has never seen before (new, dirty, incomplete data, etc.).
Artificial neural networks (ANNs) have a wide range of applications in various fields like-
1. GIS and remote sensing: ANNs are used in geographic information systems (GIS) and remote sensing applications to analyse and classify spatial data. ANNs can be trained on large datasets of satellite imagery, aerial photographs, and other spatial data to identify and classify different features, such as land cover, vegetation, and water bodies.
2. Facial and speech expression recognition: ANNs are used to classify and recognize facial expressions and speech patterns, which can be useful in fields such as psychology, human-computer interaction, natural language processing, speech-to-text transcription, and voice recognition. ANNs can be trained on large datasets of images with labelled facial expressions and audio recordings to learn to identify and classify different expressions and speech patterns.
3. Classification: ANNs are used to classify and categorize data in a wide range of applications, including image classification, text classification, and data clustering. ANNs can be trained on large datasets of labelled data to learn to identify and classify different patterns in the data. ANNs can be used to automate the process of data classification and to make predictions based on the patterns identified in the data. This is where most of the healthcare applications leverage ANNs [3].
Uses of ANN in Healthcare and Diagnostics
ANNs are used in various applications in healthcare, including disease diagnosis, drug discovery, and medical image analysis. ANNs can be trained on large datasets of medical data to learn to recognize patterns and make predictions.
1. Diagnosis and Screening: ANNs are used to analyse medical images, such as X-rays and MRIs, to detect abnormalities and diagnose diseases. Several studies have shown how neural networks can improve diagnostics and lead to more rapid decision making, which could potentially help save lives.
2. Drug Discovery: ANNs can be used to analyse large datasets of chemical compounds to identify potential drug candidates. ANNs can learn to recognize patterns in the chemical structure of compounds and predict their properties, such as their potential to bind to specific proteins or enzymes. ANNs can also be used to optimize the design of drug molecules to improve their efficacy and safety.
3. Disease Identification: ANNs can be used to analyse electronic health records (EHRs) to identify patterns and trends in patient data. ANNs can learn to recognize risk factors for certain diseases, predict patient outcomes, and identify opportunities for preventive care.
4. Personalized Medicine: ANNs can be used to develop personalized treatment plans for patients based on their genetic and medical data. ANNs can analyse large amounts of data to identify relationships between genetic and environmental factors and patient outcomes and can make predictions about which treatments are likely to be most effective for individual patients.
5. Patient Outcomes: ANNs can be used to predict patient outcomes and assess the risk of complications or disease progression. ANNs can analyse patient data, such as vital signs, lab test results, and medical history, to identify patterns that may indicate a higher or lower risk of adverse events.
6. Treatment Decisions: ANNs can be used to provide decision support for clinicians by analysing patient data and making recommendations for treatment or diagnostic testing. ANNs can assist clinicians in making more accurate and informed decisions by providing additional information and insight into patient data.
Use Case of ANN in Cancer Screening and Diagnosis
The incidence of cancer has increased in recent years, with an accompanying increase in its impact on social, physical and mental dimensions of human life. Breast cancer is the most common kind of cancer in women. The early detection, prognosis, therapeutic response prediction, and population screening is a tedious task due to the lack of specific molecular markers. The solution to these issues is to devise targeted therapeutic interventions and identify ways to improve quality of life of the patients. Risk factors include lifestyle, immunity, genetic factors, environmental factors, and underlying medical conditions. At the molecular level, activation of human epidermal growth factor receptor 2 (HER2, encoded by ERBB2) and hormonal receptors (oestrogen receptor and progesterone receptor) are responsible for progression of breast cancer. Most inherited cases of breast cancer are associated with mutation in two genes: BRCA1/2.
The current radiological methods for screening of breast cancer include mammography, ultrasound, thermography and magnetic resonance imaging (MRI), with mammography being the most widely used. Mammography is a type of X-ray imaging that uses low-dose radiation to produce images of the breast tissue.
Neural network systems can be used to automate the analysis of mammogram components, such as the parenchymal pattern and the presence of masses or calcifications and hence can analyse large amounts of data and identify patterns that may not be immediately apparent to the clinicians, thus improving the accuracy and effectiveness of healthcare interventions (Figure 1). ANNs can be used to classify mammography images and to predict the likelihood of developing breast cancer based on clinical risk factors. ANNs can analyse mammography images to identify early signs of breast cancer, such as microcalcifications and masses and also predict the risk based on risk factors such as age, family history, and lifestyle factors. They can also be used to integrate multiple diagnostic tests, such as mammography, ultrasound, and MRI, to improve accuracy and reduce false-positive rates and can improve the efficiency and speed of breast cancer screening and diagnosis (1).
Figure 1: Flow of digital mammograms
One such example of a deep learning model used for breast cancer detection using high-resolution mammogram images is discussed by Ibrokhimov and Kang (2022) (3). One of the main challenges in breast cancer diagnosis is that most models classify detected tumors/abnormalities into benign and malignant categories (binary classification). However, in the real world, the Breast Imaging Reporting and Database System (BI-RADS) score is used, which is a standard scoring system used by radiologists to report mammogram results. This scoring system doesn’t use binary classification and each score recommends a specific clinical screening routine which is important for appropriate patient treatment and follow-up screening.
Figure 2: BI-RADS classification taken from Ibrokhimov and Kang (2022) (4)
Advantages of ANNs:
The very first advantage of neural networks is that they lead to effective visual analysis. Since an artificial neural network is like that of a human’s neural network, it can perform more complex tasks and activities as compared to other machines.
Another big advantage of neural networks is that it is capable of processing unorganized data. Have you ever wondered how artificial intelligence and machine learning organize bits of data? The answer lies in the ability of neural networks. By processing, segregating, and categorizing unorganized data, ANNs can very well organize data.
The third advantage of neural networks is that their structure is adaptive in nature. This means that for whatever purpose an ANN is applied, it alters its course of the structure according to the purpose.
Another major advantage is the user-friendly interface. This is a pre-requisite for any machine or artificial equipment to be successful.
Disadvantages of ANNs:
Despite their ability to quickly adapt to changing requirements, neural networks can be a bit hefty to arrange and organize. This means that they require heavy machinery and hardware equipment.
The second demerit of neural networks is that they can often create incomplete results or outputs. Since ANNs are trained to adapt to the changing applications of neural networks, they are often left untrained for the whole process.
Another challenge of neural networks is that they are highly dependent on the data made available to them. Hence the efficiency of any neural network is directly proportional to the amount of data it receives to process.
Despite some disadvantages, the potential that ANN holds in the field of healthcare is immense and can truly help clinicians and doctors focus on the patient itself, while machines help reduce the time spent in screening, diagnosis, and treatment.
References
[1] D Bhamare,P Suryawanshi “Review on Reliable Pattern Recognition with Machine Learning Techniques” Fuzzy Information and Engineering,Taylor & Francis ,2019, VOL. 10, NO. 3, 362–377.
[2]. D Nasien1*, V Enjeslina1, M. H Adiya1, Z Baharum “Breast Cancer Prediction Using Artificial Neural Networks Back Propagation Method”,International Conference on Robotic Automation System 2021 (ICORAS 2021),Journal of Physics: Conference Series,2319 (2022).
[3]H U Sharif “Breast Cancer Detection using Artificial Neural Networks” International Journal for Research in Applied Science & Engineering Technology (IJRASET), ISSN: 2321-9653, Volume 9 Issue X Oct 2021.
[4]B Ibrokhimov,Y Kang “Two-Stage Deep Learning Method for Breast Cancer Detection Using High-Resolution Mammogram Images”,https://www.mdpi.com/2076-3417/12/9/4616.
With over 11 years of experience in data science, SAS, and research and development, Ankur has a strong background in statistical analysis, big data, and algorithm design. He has edited books on artificial intelligence, machine learning, and big data for healthcare applications.