You are here : Home > News > Determining antibiotic resistance with artificial intelligence

Scientific result | Highlight | News | Antibiotics | Artificial intelligence

Determining antibiotic resistance with artificial intelligence

Under the coordination of the MSF Foundation, researchers and engineers from the University of Évry, the CEA, the CNRS, Médecins Sans Frontières and the Bacteriology/Virology Department of the Henri-Mondor hospital (AP-HP) teamed to develop a mobile application to ease the determination of antibiotic resistance.

This application aimed at addressing a major public health issue will be freely available to medical personnel worldwide once clinically validated and CE certified.

The demonstration of the technical feasibility of their approach was published on 19 February 2021 in Nature Communications.

Published on 23 February 2021

Antimicrobial resistance: a major threat to health worldwide

The World Health Organization (WHO) has placed the growing resistance of microorganisms to antimicrobials as one of the largest healthcare challenges for the 21st century. Antimicrobial resistance could indeed surpass cancer to become the leading cause of mortality worldwide with an estimated 10 million yearly deaths, 90% of which are expected to occur in Asia and Africa due to a lack of medical means in those regions. The responsible use of antibiotics in particular has thus become primordial and with it the robust evaluation of bacterial susceptibility to them.

The international organization Médecins Sans Frontières (MSF) has been active for years in the combat against antibiotic resistance, particularly in war zones, where its physicians treat many people with multidrug resistant infections.

The most commonly-employed approach to determining bacterial resistance to antibiotics is a disk diffusion test (DDT), the results of which are summarized in an antibiogram (noting however that the terms are often used interchangeably). In a DDT, bacteria from a patient are cultured on agar growth medium in a Petri dish. A paper disk with precise concentrations of a range of antibiotics embedded in it is placed on the Petri dish as well. The antibiotics within the disk diffuse into the growth medium. The growth of bacteria susceptible to a given antibiotic will be hampered around the area where that antibiotic diffused. The resulting non-colonized area is called a zone of inhibition. The diameter of this latter is measured and compared to established standards to define bacterial susceptibility to the antibiotic. The process must respect precise rules established by experts in microbiology.

Although achievable by hand, industrialized countries benefit from automated—but costly—systems able to culture, read and interpret DDTs in a standardized fashion. The antibiogram resulting from the DDT will guide the physician's choice of an antibiotic that both treats the patient's infection efficaciously and diminishes the risk of the development of resistant bacteria.

In developing countries however, the determination of antibiotic resistance is more difficult, as was discovered by MSF physicians during the deployment of bacteriological laboratories in five resource-limited countries.

Indeed it was MSF Referent Microbiologist Nada Malou's observation of this problem after several years in the field that moved CEA Researcher Amin Madoui of Genoscope's Genomics Metabolics Laboratory (CEA/CNRS/University of Évry at Genopole) to propose a mobile application: "We needed to create a free and easy-to-use app and develop novel algorithms to enable the reading of a DDT on a smartphone."

The MSF Foundation saw therein an opportunity to bring an innovative solution to a real-world problem. Thus, in 2018, it brought together researchers from Genoscope's Genomics Metabolics Laboratory (CEA/CNRS/University of Évry), the Évry Mathematics and Modeling Laboratory (CNRS/University of Évry at Genopole) the Henri Mondor Hospital's Bacteriology/Virology Department and MSF itself to create an open-source smartphone application for all medical professionals, wherever they may be in the world, looking to read and interpret DDT results.

In 2019, the project's strong societal impact was recognized by a grant¹,²,³ from the Google AI Impact Challenge. That grant enabled human means and deep learning trials to improve the app's ability to recognize writing (the algorithm recognizes the names of all antibiotics).

Because it is a vital aspect in low-resource countries, the application is able to function without an internet connection. Users photograph their DDTs and thereafter follow the app's prompts through the analysis process. They may nonetheless intervene at any moment to verify and if needed correct the automated measures.

The application combines original algorithms, machine learning and image processing. A high-performance expert system provided by the company i2a validates data coherency and furnishes interpretations of the results. The procedure is entirely automated and provides a 98% agreement rate with the conventional gold standard of manual measurement.

Presently, the objective for the app is to ensure its performance in the resource-limited environments where MSF operates. Once completely validated, the novel tool will empower antibiotic susceptibility testing for patients everywhere and contribute to the worldwide effort for antibiotic resistance monitoring.

MSF is currently evaluating the application's clinical performance in three countries with the goal of deploying it in its laboratories before end of year 2021. Thereafter, the app will be made available to all labs in resource-limited countries in 2022.

Closing in on the end of evaluations and the beginning of deployment, MSF is calling upon all its partners involved in the battle against antibiotic resistance to join them in making the application available to the greatest possible number of laboratories in resource-limited countries across the globe. 

Top page