Translate this page into:
A mobile based antimicrobial susceptibility testing device - prototype development and validation
For correspondence: Prof Jyotsna Agarwal, Department of Microbiology, Dr Ram Manohar Lohia Institute of Medical Sciences, Lucknow, 226 010, Uttar Pradesh, India e-mail: jyotsnaagarwal.micro@gmail.com
-
Received: ,
Accepted: ,
How to cite this article: Aravindh A, Gupta A, Sen M, Das A, Agarwal J. A mobile based antimicrobial susceptibility testing device - Prototype development and validation. Indian J Med Res. 2026;164:220-6. doi: 10.25259/IJMR_214_2026
Abstract
Background and objectives
Antimicrobial resistance is a major concern due to multidrug-resistant pathogens. Advances in smartphone technology provide opportunity to digitise antimicrobial susceptibility testing (AST). This study aimed to develop and validate an in-house semi-automated smartphone application (app) for the measurement and interpretation of AST.
Methods
In this prospective validation study, an in-house app was developed using an AI-assisted coding workflow. The app uses a digital calibration algorithm that correlates pixel dimensions with millimetres (mm). AST was performed using the Kirby–Bauer disk diffusion method (KBDD) on Escherichia coli (n=48), and Staphylococcus aureus (n=15) and zone diameters were measured manually, interpreted using Clinical and Laboratory Standards Institute (CLSI) followed by analysis of the same plates using the app. S. aureus (14 drugs) required at least two Mueller–Hinton agar (MHA) plates per isolate, while E. coli (22 drugs) required at least three MHA plates per isolate.
Results
AST was performed on 1,000 antibiotic disks. Zone diameters were measured and interpreted by both methods. Categorical agreement was 90.4%. Unweighted Cohen’s Kappa was 0.836 (95% CI: 0.806–0.866). Categorical discrepancies occurred in 96 antibiotic disks. Very major, major, and minor error rates were 0%, 0.83%, and 9.2%, respectively.
Interpretation and conclusions
The app demonstrated reliable interpretation of AST, with high categorical agreement and less than 1% major error rates, supporting its safety for resistance detection. This approach offers a user-friendly alternative for busy, resource-limited settings that can be used routinely.
Keywords
Antimicrobial susceptibility testing
Disk diffusion
Drug resistance
Smartphone applications
Antimicrobial resistance (AMR) has emerged as one of the most significant threats to global public health.1 The Kirby–Bauer disk diffusion (KBDD) method, performed according to Clinical and Laboratory Standards Institute (CLSI) guidelines, is the most commonly accepted and routinely practiced method for phenotypic AST.2
Despite its widespread utility, manual measurement of zones of inhibition (ZOI) has several limitations like it is labour-intensive and subjected to inter-observer variation.3 In addition, interpretation becomes particularly challenging in isolates exhibiting indistinct or trailing (‘fuzzy’) zone edges, further increasing subjectivity in the manual reading. Manual documentation of the results and cross-referencing of zone diameters with annually updated CLSI guidelines is also time-consuming, increasing the workload of laboratory personnel, especially in a busy setup.4 Manual measurements is also prone to transcription errors.5
A variety of automated instruments based on different methodologies, including artificial intelligence, have been introduced for antimicrobial susceptibility testing (AST). The primary advantages of automation include standardisation, leading to improved accuracy, faster availability of results, and enhanced data management.6 The widespread availability of high-resolution smartphone cameras, coupled with advances in mobile computing, provides a promising opportunity to digitise AST without the need for expensive or specialised hardware. Although several digital tools for AST have been explored, the availability of low-cost, customisable smartphone-based applications remains limited. Only a few studies have evaluated smartphone-assisted approaches for disk diffusion AST interpretation in routine laboratory practice. This highlights the need for a validated, offline digital tool capable of standardising interpretation while preserving the cost-effectiveness of traditional methods.
We developed an in-house, semi-automated smartphone application (app) to assist zone diameter measurement and CLSI-based interpretation. The aim of this study was to develop and validate this app for the measurement and interpretation of AST results. The primary objective was to evaluate the accuracy and diagnostic reliability of the app for measuring and interpreting ZOI in comparison to the manual measurement method, as defined by the CLSIM100, 35th Edition (2025).7
Methods
This prospective validation study was conducted by the department of Microbiology, Dr Ram Manohar Lohia Institute of Medical Sciences, Lucknow, Uttar Pradesh, India, between November 2025 to January 2026. This study utilised routine AST plates, which were part of the standard patient care. No additional samples were collected, and no patient identifiers were recorded. The study involved analysis of anonymised laboratory data without any direct patient interaction.
Sample size and performance of AST
Bacterial identification was routinely performed according to the manufacturer’s instructions using the MALDI-TOF MS (VITEKMS, BioMérieux, Marcy-l’Étoile, France) system available in our laboratory. All clinical isolates processed during the study period were identified in-house, and isolates confirmed as Escherichia coli and Staphylococcus aureus were included in the study.
Time bound and convenience based sampling was done and 1,000 antibiotic disks were taken into analysis. Clinical isolates of all Escherichia coli and Staphylococcus aureus were consecutively included until 1,000 disks were achieved, resulting in a final sample size of 48 Escherichia coli and 15 Staphylococcus aureus isolates. Duplicate isolates from the same patient were excluded.
AST was performed using the KBDD method on Mueller–Hinton agar (MHA), in accordance with CLSI guidelines (CLSI M100, CLSI 35th Edition (2025)). Zone diameters were measured manually by using a ruler, and interpretative categories were assigned as susceptible(S), intermediate (I), or resistant (R) as per CLSI guidelines. Antibiotics for which a breakpoint was not provided in the CLSI guidelines were excluded. AST was performed using the following commercially available disks (HiMedia Laboratories Pvt. Ltd., Mumbai) in accordance with CLSI M100 guidelines ( Table I).
| Name of the microorganism | Escherichia coli | Staphylococcus aureus |
|---|---|---|
| Name of the antibiotic disk (disk content) | Ampicillin/sulbactam (10/10 µg) | Erythromycin (15 µg) |
| Piperacillin/tazobactam (100/10 µg) | Cefoxitin (30 µg) | |
| Gentamicin (10 µg) | Clindamycin (2 µg) | |
| Ciprofloxacin (5 µg) | Tetracycline (30 µg) | |
| Levofloxacin (5µg) | Doxycycline (30 µg) | |
| Trimethoprim–sulfamethoxazole (1.25/23.75 µg) | Trimethoprim–sulfamethoxazole (1.25/23.75 µg) | |
| Nitrofurantion (300 µg)* | Penicillin (10U) | |
| Doxycycline (30 µg) | Linezolid (30 µg) | |
| Cefipime (30 µg) | Gentamicin (10 µg) | |
| Meropenam (10 µg) | Ciprofloxacin (5 µg) | |
| Imipenam (10 µg) | Levofloxacin (5 µg) | |
| Imipenem (10 µg) + EDTA (750 µg) | Chloramphenicol (30 µg) | |
| Ertapenam (10 µg) | Nitrofurantoin (300 µg)* | |
| Tobramycin (10 µg) | Norfloxacin (10 µg)* | |
| Amikacin (30 µg) | ||
| Tetracycline (30 µg) | ||
| Cefoxitin (30 µg) | ||
| Fosfomycin (200 µg)* | ||
| Aztreonam (30 µg) | ||
| Ceftazidime (30 µg) | ||
| Norfloxacin (10 µg)* | ||
| Cefazolin (30 µg) |
*Urine only drugs – should be reported only on organisms isolated from the urinary tract
Quality control (QC)
To ensure the accuracy and reliability of the testing environment, QC testing was performed with each new lot of antibiotic disks and each new lot of Mueller–Hinton agar (MHA) plates using standard American Type Culture Collection (ATCC) reference strains, in accordance with CLSI M100 guidelines. Reference strains utilised were Escherichia coli ATCC 25922 and Staphylococcus aureus ATCC 25923, which were obtained from HiMedia Laboratories Pvt. Ltd., MumbaiTo assess the precision and reproducibility of the digital tool, QC runs were performed and analysed thrice for each ATCC strain. The digital measurements obtained were compared against the expected CLSI quality control ranges.7
A custom-developed, semi-automated smartphone app was used for digital measurement and interpretation of inhibition zones accessible at https://drive.google.com/uc?export=download&id=17Ax3fs1Hwh7Xq HJLVBtcc_G6MepdyrCr. The app was developed using an AI-assisted coding workflow for code optimisation, debugging, and the refinement of the measurement algorithms. No third-party commercial developers were involved, ensuring the tool’s specific nature for clinical microbiology.
Android app workflow
(i) Image pre-processing: When the user captures the image, the app utilises OpenCV-based filters to normalise the exposure and contrast of the MHA surface. This enhances the visual distinction between the bacterial lawn and the zones of inhibition (ZOI), particularly for isolates exhibiting fuzzy edges.
(ii) Spatial scaling digital calibration
To produce reliable measurements independent of the camera specifications, the software establishes a pixel-to-millimetre ratio through a manual calibration step. The user adjusts a primary calibration slider to align a digital ring with the known 6 mm diameter of the antibiotic disk. This establishes a fixed spatial scale, allowing the app to convert digital screen coordinates into precise physical measurements regardless of camera distance.
(iii) Coordinate anchoring and interactive measurement
When the user taps a disk, the app locks a digital calliper to the central coordinates of that disk, eliminating the parallax errors and off-centre biases common with manual ruler placement. The user then utilises a secondary measurement slider to expand the concentric digital ring to the ZOI edge.
(iv) Real-time guideline mapping
As the measurement slider is adjusted, the app provides an instantaneous millimetre reading and simultaneously cross-references the diameter with the preloaded CLSI M100-ED35 database within the app. This results in an immediate categorical interpretation (S/I/R), bypassing the manual clerical steps where transcription errors frequently occur5 (Figs. 1 and 2).
To assist in the accurate identification of indistinct fuzzy zone margins, the app provided multiple viewing modes, including enhanced contrast and gray scale transformation, enabling clearer visualisation of inhibition zone boundaries. Beyond individual zone measurement, the app features an integrated automated interpretation engine designed to detect complex resistance phenotypes. The software utilises a comparative algorithm that monitors specific drug-inhibitor pairs as they are added to the results list. For the identification of metallo-beta-lactamases (MBL), the app performs a real-time comparison between a plain carbapenem (e.g., imipenem) and a carbapenem-EDTA combination disk (IE) in imipenem resistant isolates. An increase in zone diameter of >5 mm with the IE disk compared to imipenem alone was interpreted as MBL producer. For Staphylococcus aureus, the app cross-references the cefoxitin (30µg) zone diameter with specific MRSA breakpoints (≤21mm), instantly labelling the isolate as methicillin-resistant where applicable. Upon completion of a plate reading, all data points, including unique AST ID, organism type, individual zone diameters, and detected resistance patterns, are consolidated into a master research log. The app automatically saves all results into a single file that can be opened in Excel. This provides a clear, permanent record that makes it easy to track resistance trends and compare the app’s accuracy against manual measurements.
In accordance with CLSI M02 guidelines, discrete colonies observed within inhibition zones were considered significant. Such colonies were visually identified during image capture, and Gram staining was performed to evaluate the colony morphology. Plates showing mixed growth or differing morphology were considered contaminated and AST was repeated. When Gram staining revealed colonies with similar morphology, they were interpreted as mutants or heteroresistant colonies, and zone diameters were measured up to the inner colony margin, where applicable.
Statistical analysis
The data were compiled using Microsoft Excel. All statistical analyses were performed using SPSS software (version 26) (Armonk, New York, USA). The accuracy of the app was evaluated by comparing its results against manual measurements, which served as the reference standard. Categorical agreement was defined as the percentage of isolates classified into the same clinical category (S, I, or R) by both methods. As a measure of agreement between the app and the manual method, we calculated the unweighted Cohen’s kappa index. The resulting Kappa values were interpreted to determine the strength of agreement, with values above 0.81 representing almost perfect agreement. This was achieved when both the app and the manual method produced the same clinical interpretation (sensitive, intermediate, or resistant). Error analysis was calculated as the rate of identical categorisation; disagreement is classified as very major, major, and minor. Very major errors (app shows sensitive, manual shows resistant), major errors (app shows resistant, manual shows sensitive), or minor errors (one method shows intermediate and the other shows s or r). To further evaluate the relationship between the two methods, Bland–Altman analysis was performed. The mean bias and standard deviation (SD) were calculated to determine the 95% limits of agreement (LoA) (mean ±1.96 x SD), establishing the clinical range of expected difference between the methods. To evaluate the inter-observer variability, zone diameters were measured independently by two observers, and the mean value was used for analysis. Inter-observer reliability was assessed using the intraclass correlation coefficient. The time required for zone measurement and interpretation by each method was also recorded.
Results
Quality control testing using ATCC reference strains Escherichia coli 25922 and Staphylococcus aureus 25923 demonstrated that all zone diameter measurements obtained manually and through the app were within the acceptable CLSI quality control ranges.
A total of 1000 antibiotic disks were tested for susceptibility on various clinical isolates of Escherichia coli (n=48) and Staphylococcus aureus (n=15). All inhibition zones were successfully measured and interpreted using both methods. The distribution of interpretative categories by organisms using the manual method is shown in Table II.
| Organism | Susceptible (S) | Intermediate (I) | Resistant (R) | Total |
|---|---|---|---|---|
| E. coli (n=48) | 390 | 88 | 351 | 829 |
| S. aureus (n=15) | 89 | 1 | 81 | 171 |
| Total | 479 | 89 | 432 | 1000 |
Categorical agreement between methods (Manual vs app)
The overall categorical agreement between the manual method and the app was 90.4% ( Table III). The resulting unweighted Cohen’s Kappa was 0.836 (95% CI: 0.806–0.866), which signifies almost perfect agreement.8
| Error type | Definition | Number of disks (n) | Percentage error (%) |
|---|---|---|---|
| Very major error | App S, Manual R | 0 | 0 |
| Major error | App R, Manual S | 4 | 0.83 |
| Minor error | Any S/R vs. I | 92 | 9.2 |
| Categorical agreement (CA) | Overall agreement between methods | 904 | 90.4 |
Very major error- denominator was the total number of R by the manual method. Major error- denominator was the total number of S by the manual method. Minor error- denominator was the total number of isolates tested
Error analysis
Among the 1,000 antibiotic disks tested, categorical discrepancies occurred in 96 antibiotic disks. The distribution of very major, major, and minor errors is summarised in Table III.
Bland–Altman analysis of zone diameter measurements
Differences between the app and the manual zone diameter measurements were evaluated by comparing paired readings. The difference within ±2 mm was observed in 82.1% (821/1,000) of the measurements. Bland–Altman analysis further demonstrated close agreement between the two methods, with a mean bias of 0.74 mm and 95% limits of agreement ranging from +4.15 mm to -2.68 mm. The mean absolute difference between the manual and the app measurements was 1.25 mm, indicating low average measurement error ( Fig. 3).
Inter-observer variability and reliability
Inter-observer variability was assessed to evaluate the reproducibility of zone diameter measurements. The mean absolute inter-observer difference was 1.41 mm for the manual method and 1.64 mm for the app. Inter-observer reliability, assessed using the intraclass correlation coefficient (ICC), was excellent for both methods. The single-measure ICC for manual measurements was 0.953 [95% confidence interval (CI): 0.922–0.972, P<0.001], while the ICC for the app measurements was 0.935 (95% CI: 0.894–0.961, P<0.001). The mean time required for the manual measurement and interpretation was 1.94 min per plate, whereas the app required 1.50 min per plate.
Beyond overall zone diameter agreement, the app was further evaluated for its ability to identify clinically significant resistance phenotypes. The app demonstrated complete concordance with the manual method in detecting methicillin resistance among Staphylococcus aureus isolates and MBL production among Escherichia coli isolates. All 10 MRSA isolates were correctly identified by the app, and all 4 MBL-producing isolates were accurately detected. These findings indicate that the app is capable of reliably recognising key resistance mechanisms of major clinical importance.
Discussion
This study evaluated a custom-developed, semi-automated smartphone app for the measurement and interpretation of AST. The app demonstrated high agreement with the manual method, indicating that the app based interpretation is clinically comparable to conventional reading. Overall, the app demonstrated a clinically acceptable error profile and reliable susceptibility interpretation. Very major errors were not observed; low rates of major error and minor error were observed. The Bland–Altman analysis demonstrated close agreement between both the measurements, with a small positive mean bias indicating that manual measurements were on average slightly higher than the app readings. Furthermore, the low mean absolute difference confirms the high accuracy of the app. Inter-observer variability analysis demonstrated good reproducibility of zone diameter measurements for both methods. The mean absolute inter-observer difference was slightly lower with the manual method compared with the app, indicating only a minimal difference between observers. The findings suggest that the semi-automated workflow can help reduce observer-dependent variability while still preserving user control over measurement adjustments. In addition, the app improved reading efficiency by reducing the time required for AST interpretation, highlighting its potential to streamline laboratory workflow without compromising measurement consistency or clinical reliability in a busy laboratory. In addition to routine susceptibility categorisation, accurate identification of clinically significant resistance phenotypes is essential for the practical utility of any AST interpretation tool. The app’s ability to identify MRSA and MBL supports its potential use in routine laboratory workflows requiring reliable detection of key resistance mechanisms.
Comparable levels of agreement have been reported for established automated disk diffusion reading systems where they evaluated the commercial ADAGIO automated reader, which reported a high categorical agreement (CA) with manual zone measurements after visual adjustment of automated readings.9 Another study evaluating a fully automatic smartphone application reported agreement levels of approximately 90% against a hospital-standard automatic reader and 98% against manual measurements, demonstrating that smartphone-based AST reading can achieve high concordance with established methods.10 The ZOI difference less than 2 mm observed in our study was similar to a previous study.4 Most discordant results were minor errors and were mainly due to zone diameters close to interpretive breakpoints, which is a known limitation of disk diffusion testing.11 Similar pattern of discrepancies has been reported in previous studies.12
A key strength of the app is its ability to provide standardised, digital assistance for disk diffusion interpretation using commonly available smartphone hardware. The offline, semi-automated workflow allows implementation without additional infrastructure, making it suitable for routine laboratory use, particularly in resource-limited settings. This study has certain limitations. The discrepant results between the app and the manual method were not verified using an MIC-based reference method such as VITEK. The evaluation was performed on a limited number of bacterial species, and resistance phenotype analysis was restricted to selected mechanisms. Image capture was performed under relatively controlled conditions, and performance under variable real-world lighting and background conditions requires further evaluation. As this was a single-centre study, the findings may not be fully generalisable to all laboratory settings. Also, the app was semi-automated, and artificial intelligence–based automation was not implemented.
In conclusion, the present study demonstrates that the semi-automated smartphone app can reliably support disk diffusion AST interpretation with high agreement to manual reading. By combining digital measurement assistance with user-guided verification, the app offers a practical and accessible tool that not only maintains clinical reliability but also reduces the time required for zone measurement and susceptibility interpretation, thereby supporting more efficient routine laboratory workflows. It also suggests that the app represents a robust and standardised tool capable of bridging the gap between manual interpretation and high-cost automated systems, particularly in resource-limited settings. Thus, the use of this app strengthens not just diagnostic but antimicrobial stewardship.
Author contributions
AA, JA, AG, MS, AD: Conception and design of study; AA, AG, MS: Acquisition of data; AA, AG, AD: Analysis and interpretation of data; JA, AG, AA, MS, AD: Intellectual content, manuscript writing. All authors have read and approved the final printed version of the manuscript.
Financial support and sponsorship
None.
Conflicts of Interest
None.
Use of Artificial Intelligence (AI)-Assisted Technology for manuscript preparation
The authors confirm that there was no use of AI-assisted technology for assisting in the writing of the manuscript and no images were manipulated using AI.
References
- World Health Organization. Global antibiotic resistance surveillance report 2025. Available from: https://www.who.int/publications/i/item/9789240116337, accessed on January 19, 2026.
- Antimicrobial susceptibility testing: A comprehensive review of currently used methods. Antibiotics (Basel).. 2022;11:427.
- [CrossRef] [PubMed] [PubMed Central] [Google Scholar]
- Recommendations of the Spanish antibiogram committee (COESANT) for in vitro susceptibility testing of antimicrobial agents by disk diffusion. Enferm Infecc Microbiol Clin (Engl Ed).. 2023;41:571-6.
- [CrossRef] [PubMed] [Google Scholar]
- Standardisation of disk diffusion results for antibiotic susceptibility testing using the sirscan automated zone reader. BMC Microbiol.. 2013;13:225.
- [CrossRef] [PubMed] [PubMed Central] [Google Scholar]
- Comparison and evaluation of Osiris and Sirscan 2000 antimicrobial susceptibility systems in the clinical microbiology laboratory. J Clin Microbiol.. 2003;41:3627-30.
- [CrossRef] [PubMed] [PubMed Central] [Google Scholar]
- The impact of laboratory automation on efficiency and accuracy in healthcare settings. Int J Community Med Public Health.. 2023;11:459-63.
- [CrossRef] [Google Scholar]
- Clinical and Laboratory Standards Institute. CLSI supplement M100. Performance standards for antimicrobial susceptibility testing. 35th ed. Available from: https://clsi.org/shop/standards/m100-ed35, accessed on January 19, 2026.
- The measurement of observer agreement for categorical data. Biometrics.. 1977;33:159-74.
- [CrossRef] [PubMed] [Google Scholar]
- Evaluation of an automated system for reading and interpreting disk diffusion antimicrobial susceptibility testing of fastidious bacteria. PLoS One.. 2016;11:e0159183.
- [CrossRef] [PubMed] [PubMed Central] [Google Scholar]
- AI-based mobile application to fight antibiotic resistance. Nat Commun.. 2021;12:1173.
- [CrossRef] [PubMed] [PubMed Central] [Google Scholar]
- Direct susceptibility testing by disk diffusion on clinical samples: A rapid and accurate tool for antibiotic stewardship. Eur J Clin Microbiol Infect Dis.. 2015;34:1207-12.
- [CrossRef] [PubMed] [PubMed Central] [Google Scholar]
- Comparison of the accuracy of disk diffusion zone diameters obtained by manual zone measurements to that by automated zone measurements to determine antimicrobial susceptibility. J Microbiol Methods.. 2008;75:177-81.
- [CrossRef] [PubMed] [Google Scholar]