The fundamental role of AI in the pharmaceutical industry

Algorithms, machine learning and deep learning are the foundations of artificial intelligence. Compared to programming, which executes predetermined rules, the machine determines a solution to a specific situation on its own. It must reproduce the functioning of a human brain and its four main cognitive capacities: learning, perceiving, understanding and acting.

Machine learning is the first training phase of artificial intelligence. Initially, the machine is fed with input data and their responses so that the algorithm can train itself and develop a prediction model. This is supervised learning. Next comes unsupervised learning or the prediction phase which is based on input data whose responses have not yet been identified. The machine must now be able to determine a solution to a given situation.

Deep learning is based on the ability of the machine to learn by itself from raw data, like a neural network. It makes artificial intelligence autonomous. Although very powerful, it is still expensive because this system needs to integrate and assimilate a considerable volume of data and hundreds of millions of images before giving reliable results.

Applied to medical diagnosis, artificial intelligence and deep learning have already demonstrated their powerful capabilities. This is the case in ophthalmology. Indeed, as early as 2018, the FDA issued for the first time an MA to a software for screening diabetic retinopathy, a complication of diabetes that affects nearly 400 million people. The clinical study showed that the IDx-DR software developed by Digital Diagnostics, after reviewing the patient’s retinal photograph, gave the correct diagnosis 9 times out of 10. In just a few minutes, the program was able to complete the diagnosis and deliver an interpretation report as well as treatment recommendations. A major advance for early detection of this disease and avoid blindness.

Oncology is not left out. Partnerships are multiplying between health institutes and specialized companies to develop AI approaches applied to personalized medicine. The results are encouraging and offer hope for early detection of certain types of cancer. At the end of 2021, the U.S. regulatory agency granted marketing authorization to Paige Prostate, an AI-based pathology diagnostic solution specifically designed to diagnose prostate cancer. In the clinical study, pathologists saw a 70% reduction in false negative diagnoses and a 24% reduction in false positives [1]. In addition to the U.S., the diagnostic software is now approved for use in hospitals and laboratories in the European Economic Area, Switzerland and the United Kingdom, thanks to its CE mark.

Assisted by connected objects, telemedicine is now essential to ensure optimal real-time monitoring of patients suffering from pathologies such as kidney failure, sleep apnea, cancer, heart failure or even patients with transplants. In 2020, Moovcare led the way by becoming the first reimbursed digital therapy in France. Six years of R&D and proven efficacy during phase II and III clinical trials have been crowned with success, with CE certification as a class I medical device. Developed by the connected health company Sivan, Moovcare is a digital platform dedicated to the follow-up of lung cancer patients. Available on medical prescription, a simple questionnaire filled out each week by the patient is enough for the algorithm to detect a recurrence or complications, and then send the results to the prescribing physician in charge of the patient.

Some more complex connected objects go further than simple monitoring. They also provide automatic administration of treatment. In the field of diabetes, the solution developed by Diabeloop is a perfect example. It is based on an algorithm connected to a blood glucose monitoring device and an insulin pump installed on the patient. A glucose level measurement is transmitted every 5 minutes to the algorithm. The algorithm calculates the right dose of insulin to administer and automates the injection of the product after analyzing the data in real time and taking into account the patient’s personal data (history of the disease, physical activity, meals). A hybrid closed-loop system that has already convinced Terumo and Roche.