Machine learning has received a lot of attention because of its potential usage in medicine. How exactly does machine learning help clinicians and scientists in their daily work?
The idea of improving medicine through computation is as old as digital computers. In the 1960s, scientists used computers to diagnose blood diseases. This is just one of many pioneering examples in this field. Nowadays, computers learn from past experience – this is what is called ‘machine learning’ in the artificial intelligence (AI) domain. These results can be used to teach clinicians and medical researchers new methods of diagnosing and treating diseases and handling patients.
In the clinic
According to Pearse KEANE, consultant ophthalmologist at Moorfields Eye Hospital NHS Foundation Trust in the UK, machine learning is promising in the clinic but there are still many challenges. He says that there’s a big gap between proving a concept in a research paper and the actual usage of the machine learning system in the real time. Keane adds that machine learning has the potential to transform healthcare. However, it is difficult to go ‘from code into clinic’.
1. Improved prognostics
Machine learning can be used by clinicians to help predict the future. Cancer is the most obvious application. One example is the machine learning-based tool developed by an international team of scientists that analyses the prognosis for patients with stage III colon carcinoma. The group stated that these results could provide “crucial information to aid treatment planning”. John Halamka (president of the Mayo Clinic Platform) and his colleagues suggested machine learning could improve clinicians’ ability to predict the outcome of patients with COVID-19. The future of work with prognosis is bright, as machine language has been used in clinical diagnosis.
2. Patient monitoring
Traditional medicine has seen physicians only contact patients when symptoms present. Sometimes, this is not until a patient becomes seriously ill. Ali Rezai, the John D. Rockefeller IV chair in neuroscience at West Virginia University, says: “This is slowly changing. Machine learning and artificial intelligence models will one day be able to continuously monitor a person’s health”. Rezai explains that two of the most popular AI systems are available in devices such as the Apple Watch and the Kardia Alivecor. These devices can detect arrhythmias, send alerts and notifications to patients via their smartphone apps. AI is still not fully integrated in the clinical flow. However, it will have a significant impact on cardiology, neurosciences, and cancer. It can help to stratify and profile patients and enable more proactive management and care.
3. Developing diagnostics
Computing can be used to analyze images in a variety of applications, including military and medical. Nqoba Stabze, an imaging and clinical expert, and Dineo Mpanya, a South African hospital professor at the Charlotte Maxeke Johannesburg Academic Hospital, in Johannesburg, collaborated to discuss the interpretations of machine learning on MRI image datasets such as chest radiographs. Scientists note that subtypes of machine-learning, such convolutional neural network, can detect subtle changes in chest radiographs and, in some cases, diagnose conditions like pneumonia with a higher accuracy than clinicians. Machine-learning algorithms mimic human cognitive processes when making decisions, unlike traditional statistical methods that are based on inferences from the studied population.
Check out https://www.medicaldata.cloud/ for more MRI image datasets to enhance the quality of laboratory and hospital research.
The US Food and Drug Administration approved the first IDx-DR AI-based diagnostic in April 2018. It analyzed retinal images and detected diabetic retinopathy. Machine learning will soon be used to treat many other medical conditions, including cardiology and neurodegenerative disorders.
4. Collaborations are required
Machine learning is a promising tool for collaboration, perhaps more than any other medical technology. In fact, it is what makes machine learning-based applications so valuable. Maria Littmann (a doctoral candidate in bioinformatics from Technical University Munich) and her collaborators discovered that, when they analysed 250 articles about machine learning applications in medicine or biology. These scientists discovered 73% of machine-learning applications were the result of interdisciplinary collaborations between computational scientists, medical experts, and biologists in those articles.
In the laboratory
In general, computers help greatly in the majority of clinical research areas. AI-based methods offer even more potential applications for researchers. Machine learning methods have already had an impact on many areas of clinical research though they are not widely used yet.
5. Big data
Pearse Keane also says that a large dataset meant hundreds of patients in the past. “As a result of machine learning infrastructure, we are now able to accumulate much, much larger datasets,” Keane says. Keane studies the most common reason for blindness in Europe, the USA and elsewhere. Keane anticipates that in the next decade they will be doing clinical trials using images from all patients diagnosed with age-related macular degeneration, and that the quantity of patients will reach up to hundreds and thousands of patients per year.
6. Recruiting patients
One of the most important elements in medical research is clinical trials. However, recruiting patients can be a little bit tricky. Mira Desai, a pharmacologist at the Nootan Research Centre in India stated that participant enrollment issues are ones among the main reasons for trial terminations. In this case, medical researchers could use machine learning to solve the problem. A group of scientists from Australia’s Commonwealth Scientific and Industrial Research Organisation developed a machine learning technique that analyzes medical records in order to identify people who are suitable for specific trials. This is an example of how machine learning is only getting started in clinical trials.
7. Hypothesis testing
Predicting the outcome of a given scenario in medical research is difficult. David Watson, a doctoral candidate at Oxford Internet Institute of Oxford and founder of the Digital Ethics Lab, says statistical models are the best way for structure to be revealed and outcomes to be predicted. He says that it is not always clear how to do it with clinical data alone, whereas clinicians often believe that genomic information can solve the problem. However, it is still possible to combine information from clinicians with data-science tools, such as machine learning, to develop a hypothesis, model it and adjust it. Then, replicate the process iteratively. Watson states that this requires close collaboration between data scientists and clinicians, who have different idea of the problems. However, Watson believes that good research involves bringing together people from different niches to solve difficult problems.
8. Reconstructing diseases
Colin Hill, CEO of GNS Healthcare and cofounder, says that machine learning, combined with multi-modal data sets and nearly unlimited computing power, allows clinical researchers to “reconstruct disease’s underlying mechanisms.” Gemini, GNS Healthcare’s AI-driven simulation platform, for example, provides a computer model that simulates multiple myeloma progression and drug responses. Hill explained that this model “harnesses causal machine learning and simulation, in-depth clinical data and molecular patient information to allow pharma companies simulate drug responses at the individual patient level.”
Experts in different fields have different perspectives and different methods of analyzing data. These collaborations will result in larger datasets. Hill states that machine learning has the potential to have a greater impact on clinical research as the volume of multimodal data increases. He says that we have only scratched the surface. The impact of machine learning fueled by the right data can transform the development of new, seminal medicines and improve their usage while treating patients.