Machine learning is employed extensively in healthcare, prediction, diagnosis, and as a technique of establishing priority. Artificial intelligence is widely used in the medical industry. There are a variety of tools in the disciplines of obstetrics and childcare that use machine learning techniques. The goal of the current chapter is to examine current research and development views that employ machine learning approaches to identify different complications during delivery. The common complications such as gestational diabetes mellitus, preeclampsia, stillbirth, depression and anxiety, preterm labor, high blood pressure, miscarriage were explored in this chapter. It investigated a synthesized picture of the features utilized, the types of features, the data sources, and its characteristics; it analyzed the adopted machine learning algorithms and their performances; and it gave a summary of the features employed. Eventually, the results of this review research helped to create a conceptual framework for improving the maternal healthcare system based on machine learning.
TopIntroduction
Machine learning (ML) is increasingly employed in health care, prediction, diagnosis, and as a technique of establishing priority. The World Health Organization (WHO) estimates that 800 women worldwide pass away every day from preventable diseases associated with pregnancy's inherent dangers (Bertini et al., 2022). The vast majority of these fatalities (94%) happened in areas with little resources, and the bulk of them could have been avoided. Artificial Intelligence (AI) is widely used in the medical industry. There are a number of instruments in the disciplines of obstetrics and childcare that use ML techniques. AI can assist professionals in decision-making, reduce medical errors, improve the accuracy of the interpretation of various diagnoses, and reduce the workload to which they are exposed. The goal of the current review is to provide an overview of ML methods for predicting complications in pregnancy.
Due to the enormous growth of both structured and unstructured data, Big data has made ML indispensable because it is impossible to handle this data using other approaches. Huge data helps ML systems to find previously unidentified patterns, stimulating the decision-making process. The field of ML is where computers are trained to behave similarly to humans. The utilisation of data and algorithms is emphasised. The ML technique involves handling a vast amount of data, training, and creating a machine learning model, as well as training that model to increase accuracy.
ML relies on human interaction with the raw data to enable machines or models to learn. The data may or may not be labelled. The ML model creates an approximation of a pattern based on this data. The accuracy of the estimation is then determined by comparing it to the known answer, or the labelled data. The model then makes an attempt to fit the estimation to the known data points to increase accuracy. This is how the ML method trains and develops the models that aid in the machine's imitation of human behaviour.
AI's branch of ML is a subfield of computer science. These methods enable the inference of significant relationships between data elements from disparate data sets that would otherwise be challenging to correlate. These techniques make it possible to infer meaningful correlations between data pieces from various data sets that would be difficult to correlate otherwise.
A final model fit on the training data set is evaluated objectively using the test data set. The train-test split or cross-validation procedures are typically used to validate ML models. A training data set, or collection of instances used to fit the model parameters, is typically used to initialise models. Both parameter estimation and variable selection may be involved in model fitting.
By splitting the data into k-folds, each fold is divided into two segments: one used to learn or train a model and one used to validate the model. This statistical technique is known as cross-validation, and it is used to evaluate and compare learning algorithms. The training and validation sets must be crossed in successive rounds in a conventional cross-validation. Figure 1 depict the general workflow for predicting pregnancy complications using ML algorithms.
Figure 1. Workflow for predicting pregnancy complications using ML algorithms
Source: Espinosa et al. (2021) The recent availability of high-throughput, molecular-level data from genomic, transcriptomic, proteomic, metabolomic, and single-cell immunological measurements, combined with advanced computational and statistical tools, has enabled analyses of these large and detailed datasets, as well as the integration of biological and nonbiological biomarkers. Such integrated techniques can provide more specific and generalizable signatures of pregnancy-related pathologies, allowing for more precise inferences about the diversity and multiplicity of causes of pregnancy-related pathologies.