The extraordinary abilities of artificial intelligence in terms of healthcare has always fed the imagination of science fiction writers, including in the cinema, where it regularly takes centre stage.

From Star Trek's Tricorder to the Med-Pods of Alien and Elysium, the film world is rife with futuristic devices that scan the body in a few seconds, pinpoint diseases and even cure them on the spot.

We aren't there yet, but the advances in healthcare made possible by AI are spectacular and are leading towards an increasingly individualized medicine that is both predictive and preventive. 

AI and computer vision pave the way to cutting-edge predictive medicine

In a study published in 2016, Frost & Sullivan announced that the artificial intelligence market in health care could reach $6.6 billion by 2021, representing growth of 40%. The firm also emphasises that AI, through its ability to foster the improved accessibility, relevance and usability of data, is now revolutionising the medical world, particularly in terms of diagnosis.

In this respect, computer vision technologies are among the most promising AI branches in healthcare. Algorithms used in medical analysis and imaging (radiology, MRI, ultrasound, etc.) can recognize lesions, abnormalities and the small signs characteristic of a pathology, thus helping to establish extremely reliable, accurate and rapid diagnoses.

Approved in April this year by the US Food and Drug Administration, the IDx-D software developed by the start-up IDx analyses images of the retina, for example, to detect cases of diabetic retinopathy in diabetes patients – with a success rate of over 89%. This is the first time the American health authorities have granted a marketing authorisation for a device entirely piloted by artificial intelligence, which can do without a clinician's supervision to interpret the results.

In May this year, the review Annals of Oncology published the results of a study by a French/German/American team of researchers on skin cancer detection using an artificial intelligence system. Out of 100,000 photos, AI was able to identify 95% of melanomas, compared to the 89% pinpointed by dermatologists.

And that's just a quick glance at the possible applications of artificial intelligence in the medical world. Many experts agree that in the short-term, machine learning will be an essential component of medical practice, and one particularly valuable in assisting diagnosis and prescriptions through the analysis of images analysis, medical reports and biological findings.

Training ML algorithms in healthcare

Dominated by an evidence-based and then data-driven approach, today's medicine is moving towards a "model-driven" approach with the growing use of machine-learning algorithms. To optimise doctors' diagnostic abilities and make them reliable, this approach is designed to generate learning models driven by data that are annotated and qualified in line with the diagnostic goals in view.

For example, the machine-learning algorithms used to diagnose breast cancer need to learn to recognise cancerous lesions and differentiate them from benign masses like cysts, vascular  malformations and tissue distortions. This necessitates the preparation and annotation of current data, during which the attributes and characteristics of pathological lesions are defined so that the algorithm can subsequently recognise them. As emphasised by Dr Robert Rowley, co-founder of Flow Health: "A machine-learning algorithm simply looks at data that already exists, and its precision and usefulness is a function of the data it has learned from."

Doctors thus play a crucial role here: their knowledge and experience make it possible to precisely and accurately annotate the data sets used to train an algorithm, guaranteeing its reliability. This is particularly critical in a sector where human lives are at stake: if a training data set contain errors, the consequences can be fatal for a patient. Hence the crucial importance of adopting good practices in terms of data preparation.

The issue of the collection and confidentiality of medical data

However, there is still the question of collecting medical data, with all the ethical and regulatory problems entailed by their exploitation for machine learning purposes. Because it may seem obvious, but AI needs data! And that's a big problem in France, where the healthcare sector is still lagging behind in terms of data-sharing.

And yet numerous initiatives are clearly moving towards a more open, collaborative and finally more effective approach. For example, Cédric Villani's French Report on artificial intelligence, made public in March 2018, predicts the replacement of the SNDS (national health data system) with a "data-sharing platform for research and innovation in healthcare" and the introduction of "more fluid procedures for accessing data" to develop the potential of AI in healthcare.

Initiatives are also emerging from private sector, such as the start-up Kidner, which aims to improve access to anonymous medical information on transplants via the blockchain. Or the Epidemium open science programme, where the idea is to draw on open data and an open community of transdisciplinary contributors to foster innovation in cancer research.

In the end, perhaps we are not that far from Hollywood, but all the prospects offered by AI in favour of cutting-edge predictive medicine, promising as they are, rely on the volume, nature, quality and relevance of data available to data scientists to train their algorithms properly, and obtain sufficient accuracy to meet sometimes vital objectives.

Originally published on July 09, 2018 Topics: Machine Learning Deep Learning HealthTech


You may also like: