# P-1996. Performance Evaluation of an Artificial-Intelligence-Driven Single-Cell Imaging Platform for Rapid Phenotypic Antimicrobial Susceptibility Testing

**Authors:** Dong Woo Shin, Yunsang Choi, Kyunghwa Lee, Seong Jin Choi, Song Mi Moon, Eu Suk Kim, Hong-Bin Kim, Jeong Su Park, Kyoung-Ho Song

PMC · DOI: 10.1093/ofid/ofaf695.2161 · Open Forum Infectious Diseases · 2026-01-11

## TL;DR

A new AI-based imaging platform called PhAST can rapidly test antibiotic susceptibility in bacteria from blood cultures within 90 minutes, showing high accuracy.

## Contribution

The PhAST assay introduces a rapid, AI-driven single-cell imaging method for phenotypic antimicrobial susceptibility testing directly from blood cultures.

## Key findings

- The PhAST assay achieved 95.1% categorical agreement with reference methods for Enterobacterales.
- The assay delivered results in 117 minutes on average, significantly faster than conventional methods.
- Performance was comparable to commercial systems like VITEK2 with low error rates.

## Abstract

Conventional microbiological methods require several days from specimen receipt to organism identification and antimicrobial susceptibility testing (AST), delaying appropriate therapy. We evaluated an early prototype of the PhAST assay, a novel artificial-intelligence (AI)-driven single-cell imaging platform for rapid phenotypic AST.

The prototype PhAST assay was developed to provide bacterial identification at the group level defined by the Clinical and Laboratory Standards Institute, and phenotypic AST results were obtained in ninety minutes, directly from a positive blood culture. Susceptibility was determined by AI-based quantitative analysis of single-cell phenotypes from images and videos of antibiotic-exposed and untreated cells (Figure 1). The assay was performed on positive blood culture broth samples flagged by a routine blood culture system obtained from the clinical laboratory. Only samples containing Gram-negative bacteria were tested. We also measured the time to result (TTR) of the assay. Broth microdilution (BMD) was performed in triplicate as the reference method. Results from the commercial automated AST system (VITEK2) were also assessed against BMD.

A total of 122 Enterobacterales, including 69 (56.6%) Escherichia coli, 37 (30.3%) Klebsiella pneumoniae, and 6 (4.9%) Enterobacter cloacae complex, were identified. Resistance to third-generation cephalosporins was observed in 54 (44.3%) of the isolates, based on BMD results. The prototype PhAST assay showed a mean (± standard deviation) TTR of 117 (± 3) minutes. The assay achieved favorable results, with a categorical agreement (CA) of 95.1%, and very major error (VME), major error (ME), and minor error (mE) rates of 0.2%, 1.1%, and 3.6%, respectively (Table 1). The commercial AST system exhibited CA, ME, and mE rates of 93.9%, 0.4%, and 3.9%, respectively, with a slightly high VME rate of 1.9%.

The prototype PhAST assay provided acceptable performance for Enterobacterales within two hours of blood culture positivity. This rapid phenotypic AST assay has significant potential to support timely clinical decision-making. Further expansion to additional organisms and antibiotics could enhance its clinical utility.

Jeong Su Park, M.D. / Ph.D., PhAST Corp.: Advisor/Consultant|PhAST Corp.: Grant/Research Support Kyoung-Ho Song, MD, PhD, PhAST Corp.: Advisor/Consultant|PhAST Corp.: Grant/Research Support

## Linked entities

- **Species:** Enterobacterales (taxon 91347), Escherichia coli (taxon 562), Klebsiella pneumoniae (taxon 573), Enterobacter cloacae complex (taxon 354276)

## Figures

2 figures with captions in the complete paper: https://tomesphere.com/paper/PMC12793368/full.md

---
Source: https://tomesphere.com/paper/PMC12793368