Context
AMR is rising globally, with especially high impact in low- and middle-income countries LMICs, where diagnostic capacity is limited, and microbiologists are scarce. Only 1-2% of clinical laboratories in Sub-Saharan Africa perform bacteriology testing, despite the region having the highest AMR-associated mortality. A major gap is the shortage of trained personnel and inconsistent or inaccurate antimicrobial susceptibility testing (AST) interpretation practices. This poses a threat not only for patients but also at the global level due to the rapid spread of resistant organisms.
Background
Antibiogo, a free mobile application developed by Médecins Sans Frontières (MSF), aims to support laboratory technicians, allowing for accurate AST interpretation where microbiologists are not available, thereby addressing:
- Poor access to AMR diagnostic expertise
- Non-standardised AST reading and interpretation
- Need for scalable, affordable diagnostic tools in LMICs
Earlier field evaluations in Mali, Senegal, Kenya, Laos, and Jordan demonstrated the feasibility and accuracy of Antibiogo under operational conditions, supporting its CE-marking under the European In Vitro Diagnostic Medical Devices Directive (IVDD). To establish analytical and interpretive validity under standardised conditions, an independent evaluation of the app was supported by ICARS and performed by the EUCAST Development Laboratory (EDL).
How does Antibiogo work?
When installed on a smartphone, laboratory technicians can take a photo of prepared disk diffusion AST plates following a standardised protocol, and upload them using the app. The app then uses image recognition software combined with AI to identify the antibiotic disks on the plate. If the software does not detect them automatically, the technician can enter them manually. The app then provides a semi-automatic measurement of the inhibition zone diameter (IZD) for each of the antibiotic disks on the agar plate, which can be adjusted (if needed) and approved by the technician. Once all zone diameters have been approved, the data is sent to the second software component: the expert system. The expert system applies the latest CLSI/EUCAST breakpoints and interpretive criteria and translates IZDs into susceptibility categories (susceptible: S; susceptible, increased exposure: I; resistant: R) for each of the antibiotics. Results are then generated with interpretive comments to the laboratory technicians and clinicians.
Project overview
The independent evaluation had two objectives:
- Examine the accuracy of Antibiogo’s measurement of IZDs: IZDs from different bacterial-antibiotic combinations were measured both using the Antibiogo app and manually by technicians at the EUCAST Development Laboratory according to standard procedures. A single trained technician performed Antibiogo readings, capturing images, applying the semi-automatic detection, and adjusting the IZDs as needed. This allowed a direct comparison between manual measurements and both the initial automated and the technician-adjusted Antibiogo measurements.
- Evaluate the Expert System: This feature of the app acts like a microbiologist by applying EUCAST/CLSI guidelines, flagging problematic results, and suggesting likely resistance mechanisms. To test this, the team used both the results obtained above as well as applied it on bacteria with well-known resistance traits. A laboratory technician manually measured IZDs, and a senior microbiologist trained in AMR mechanisms performed Antibiogo readings. The system’s output was then compared to the clinical microbiologist’s interpretation, considered the gold standard.
Key results
The independent evaluation results showed that Antibiogo is a highly accurate, easy-to-use tool that has high potential for supporting non-expert laboratory staff in resource-limited settings in producing reliable AST results.
Zone measurement accuracy: Adjusted zone measurements showed categorical concordance in 96.4%. Almost 88% of measurements were within 2 mm of the manual reading. These results suggest that the app is accurate, and especially so when a technician reviews the measurement. The evaluation also showed that presently there is a need for manual adjustment of zones proposed by the app as there was only categorical agreement in 69.6% of the initial proposed IZDs by the app.
Categorical agreement:
There was a 99.1% overall categorical agreement between Antibiogo’s technician-adjusted IZDs and the manual expert interpretation.
This means that the app’s interpretation is similar to the conclusions a trained microbiologist would draw.
Usability:
Reading a plate with Antibiogo took about 3 minutes on average, whilst for bacteria requiring two plates took around 5 minutes. This is roughly twice as long as manual reading for an experienced technician.
Whilst the app works well, it could slow down experienced users. However, it could greatly help less experienced technicians who would otherwise struggle to interpret results.
Physical documentation and verification: By storing photos, the app provides physical documentation of interpreted results, the app is opening possibilities for remote microbiologist verification and trust building between clinicians and the laboratory.
Next steps
The Antibiogo app has strong potential to support ICARS projects by strengthening capacity in under-resourced settings and improving antimicrobial stewardship through better diagnostic accuracy. It could be integrated into the ICARS-supported Centre of Excellence project, with training delivered by the regional expert via remote consultations to ensure effective use by laboratory staff.
Following the evaluation results, MSF plans to extend the use of Antibiogo to public health facilities in collaboration with Ministries of Health in countries such as Burkina Faso, Côte d’Ivoire, Benin, and Niger. To sustain and accelerate the global expansion, MSF are seeking partnerships with AMR stakeholders who can support the scale-up and provide user feedback for ongoing improvements.
