Caption Health
June 1, 20221956-1961
June 1, 2022Machine Learning
There are several types of Machine learning algorithms organized into a taxonomy based on the desired outcome of the algorithm...
Algorithms created for Machine Learning (ML) are categorized by taxonomy based on the desired outcome. The most common types of algorithms are supervised learning, unsupervised learning, semi-supervised learning, and reinforced learning.(1)
Supervised learning implies that each algorithm generates a function that maps given inputs to desired outputs. The classification problem is a standard scenario for supervised learning. The learner needs to acquire knowledge to approximate the behavior of a given function that classifies a vector into one of several classes by comparing several input-output examples of the function.(1)
Unsupervised learning is used for algorithms to model a set of inputs into outputs while labeled examples are not available.
Semi-supervised learning combines labeled and unlabeled examples to lead the algorithm to an appropriate function or classifier.
Reinforcement learning leads the algorithm to learn how to act given an observation of the world. Every action has an effect on the environment, and the environment delivers feedback which is the guide for the learning algorithm.(1)
ML algorithms of relevance to medicine
Several Machine Learning algorithms are in use; a few common ones are discussed below due to their regular use within the literature and their relevance to research in many fields, including medicine.
Support Vector Machine (SVM)
A support vector machine is a supervised learning algorithm designed to split data into categories. ‘Support vector’ refers to the margin used by the algorithm to support its decision of whether data falls into a category or not (Fig. 1). Researchers use ‘kernels,’ mathematical tools that adjust data to make it easier to separate it into categories. (2)
This process may seem straightforward for 2D and 3D data sets, but the marvel of SVM is that it can be used to categorize complex data sets with several variables or dimensions. Since SVM is extremely versatile, it has been used to process a variety of data types, from mammograms being classified by having microcalcifications or not to classifying tissue and cell types based on genetic microarray expression data.(2)
A support vector machine is a supervised learning algorithm designed to split data into categories. ‘Support vector’ refers to the margin used by the algorithm to support its decision of whether data falls into a category or not (Fig. 1). Researchers use ‘kernels,’ mathematical tools that adjust data to make it easier to separate it into categories. (2)
This process may seem straightforward for 2D and 3D data sets, but the marvel of SVM is that it can be used to categorize complex data sets with several variables or dimensions. Since SVM is extremely versatile, it has been used to process a variety of data types, from mammograms being classified by having microcalcifications or not to classifying tissue and cell types based on genetic microarray expression data.(2)
Neural Networks (NN)
The neural network (NN), usually called artificial neural network (ANN), is a concept derived from the biological function of neurons, cell-like structures in the brain. Thus, in order to be able to understand NNs, one must understand the structure and function of neurons.(3)
A neuron has mainly four parts: the dendrites, nucleus, soma, and axon. The dendrites receive electrical signals, and the soma processes the signals. The output of the process is carried outward by the axon reaching the dendrite terminals of another neuron. The nucleus is where the irreplaceable information for the correct functioning of the neuron is kept.(3)
A neural network is the interconnection of neurons where electrical impulses travel around the brain. An artificial neural network algorithm is taught to behave the same way; it has three layers: the input layer that receives the input (dendrites), the hidden layer that processes the signal (soma), and the output layer that sends the output (axon) to be used in another process (next neuron dendrites). ANNs can be trained using supervised, unsupervised, or reinforcement learning.(3)
The neural network (NN), usually called artificial neural network (ANN), is a concept derived from the biological function of neurons, cell-like structures in the brain. Thus, in order to be able to understand NNs, one must understand the structure and function of neurons.(3)
A neuron has mainly four parts: the dendrites, nucleus, soma, and axon. The dendrites receive electrical signals, and the soma processes the signals. The output of the process is carried outward by the axon reaching the dendrite terminals of another neuron. The nucleus is where the irreplaceable information for the correct functioning of the neuron is kept.(3)
A neural network is the interconnection of neurons where electrical impulses travel around the brain. An artificial neural network algorithm is taught to behave the same way; it has three layers: the input layer that receives the input (dendrites), the hidden layer that processes the signal (soma), and the output layer that sends the output (axon) to be used in another process (next neuron dendrites). ANNs can be trained using supervised, unsupervised, or reinforcement learning.(3)
Conclusion
Machine learning is a continuum of the merging of computer science and statistics. It may seem like the next wave in advancing modern health care, but it has already arrived, and it is being used in real-world applications with success in many specialties of medicine. (2)
Accuracy is one of the main focuses of most algorithms since human lives are and will continue to depend on them. Studies have shown that SVM provides improved accuracy of 94.60% for the detection of heart disease. The feed-forward neural network offers 98% accuracy in correctly classifying hepatitis. (4)
Physicians should become more familiar with the basic concepts of ML, and their potential applications, to embrace the increasing integration of AI and ML into modern medicine. (2)