Cancer is defined as a heterogeneous disease consisting of many different subtypes. Early diagnosis and prognosis of the type of cancer have become a must in cancer research, as they facilitate the subsequent clinical management of patients. The importance of classifying cancer patients into high- or low-risk clusters has led many research groups, from the biomedical to the bioinformatics field, to investigate on the application of machine learning methods. Therefore, this technology has been used to model disease progression and treatment. Furthermore, the ability of machine learning techniques to detect key features from complex data sets reveals how relevant their use in this field is.
Although it is clear that the use of ML methods can improve our understanding of cancer progression, an appropriate level of validation is required for these methods to be considered useful in everyday clinical practice.
In this JOurnal article we will discuss how machine learning applied to medicine can make an essential contribution to the accurate collection of data for diagnosis in the field of oncology.
MACHINE LEARNING IN CANCER DIAGNOSIS
Machine learning (ML) is the teaching of actions and activities to computers and robots. This provides the healthcare sector with a wealth of data to improve its performance and efficiency.
Over the last decade, cancer research has made huge strides in broadening its horizons. Scientists have over time adopted different methods, such as early-stage screening, in order to find different types of cancer before they cause lethal symptoms. In addition, they have developed new strategies for early prediction of outputs related to cancer treatment. The results from machine learning have been made available to allow the medical research community to study the outputs and identify new solutions.
However, accurately predicting the outcome of a disease remains one of the most challenging tasks for doctors. This task is nowadays facilitated by taking advantage of the results derived from ML methods. These techniques can identify the most significant correlations, extracting information from complex datasets, which allow future outcomes of cancer types to be predicted more effectively.
Several studies have been aimed at understanding new methodologies other than those adopted previously, based on strategies that could enable the early detection of cancer and its prognosis. In particular, these studies describe approaches related to the profiling of circulating miRNAs that have proven to be a promising class for cancer detection and identification. However, these methods currently suffer from low efficacy in early-stage screening and difficulty in distinguishing benign from malignant tumours.
MACHINE LEARNING AND DEEP LEARNING IN THE FIGHT AGAINST CANCER
AI technologies include different methods of data analysis. On the one hand, machine learning uses data that has been preprocessed and makes predictions based on what the AI learns. Deep learning, on the other, can identify complex patterns directly from raw data and is used to identify cell nuclei in huge data sets.
Deep learning has its origins in the 1940s, when scientists built a computer model that was organised in interconnected layers, like neurons in the human brain. Decades later, researchers taught these ‘neural networks’ to recognise shapes, words and numbers, but it was only a few years ago that deep learning began to expand into the medical and biological sectors.
Deep learning has been applied to many areas of health care, including imaging diagnosis, hospitalisation prediction, drug design, cancer and stromal cell classification, medical care, etc. Cancer prognosis consists of estimating the fate of cancer, the probability of cancer recurrence and progression, and providing an estimate of survival for patients. Therefore, the accuracy of cancer prognosis prediction is of great benefit to the clinical management of patients with cancer.
The improvement of translational biomedical research and the application of advanced statistical analysis and machine learning methods, through machine learning and deep learning, are the driving forces to improve the prediction of cancer prognosis. Cost reductions in large-scale next-generation sequencing and the availability of such data through open-source databases (e.g. TCGA and GEO databases) offer opportunities to build more powerful and accurate models.
Deep learning requires less data engineering and achieves accurate prediction when working with large amounts of data. Deep learning in cancer prognosis has been shown to be equivalent or better than current approaches, such as Cox-PH, a regression model commonly used statistically in medical research to investigate the association between patient survival time and one or more predictor variables. With the explosion of multi-omics data, including genomic data, transcriptomic data and clinical information in cancer studies, deep learning is leading to an exponential improvement in cancer prognosis.
Artificial intelligence and e-health have made huge advances in the medical sector. The use of increasingly intelligent automations, which enable the acquisition of crucial information from processing vast amounts of data, is a sign that digitalisation in healthcare is bearing fruit. AI is a great opportunity, but in order for this to work, a major digital transformation plan needs to be initiated in healthcare.