Digital Image Analysis for the Detection of Group B Streptococcus from ChromID Strepto B Medium Using PhenoMatrix Algorithms

Justin Baker, Karen Timm, Matthew L. Faron, Nathan A. Ledeboer, Karissa CulbreathJournal of Clinical Microbiology 2020 · DOI

TL;DR

This paper introduces an AI-powered digital image analysis system (PhenoMatrix CDM) for detecting Group B Streptococcus (GBS) on chromogenic agar. The system significantly improves upon manual culture review, achieving near-perfect sensitivity and identifying GBS cases missed by human eyes, making it a promising tool for routine clinical microbiology.

What Problem Does This Paper Address?

Group B Streptococcus (GBS) is a major cause of serious illness and death in newborns. While culture-based methods are the gold standard for detecting GBS colonization in pregnant women, they can be time-consuming and sometimes lack the sensitivity of molecular tests like NAAT (nucleic acid amplification testing). NAATs are faster and more sensitive but come with a higher cost. This creates a challenge for laboratories: balancing the need for accurate, sensitive GBS detection with cost-effectiveness and workflow efficiency. The paper aims to bridge this gap by enhancing the performance of chromogenic culture methods using AI-driven image analysis.

Methods

The core innovation here is the use of Copan's PhenoMatrix (PM) software, specifically its Chromogenic Detection Module (CDM). This AI-powered software analyzes digital images of bacterial cultures grown on ChromID Strepto B medium. The PM CDM is designed to:

  • Recognize and differentiate colony colors: This is crucial for chromogenic media, where specific colors indicate the presence of target organisms.
  • Presort culture plates: It categorizes plates as "negative for GBS" or "potential positive for GBS," streamlining the workflow for technologists.
  • Flag presumptive GBS colonies: The algorithm identifies colonies that exhibit characteristics consistent with GBS.

The study compared the performance of this AI-driven image analysis (PM CDM) against two benchmarks:

  1. Manual culture review: Technologists visually inspecting digital images of the cultures.
  2. Molecular detection (NAAT): Using BD MAX GBS and Cepheid Xpert GBS assays as a reference standard.

The study also utilized the WASP automated specimen processing system and a smart incubator to standardize culture conditions and image capture.

Datasets

The study used a dataset of 676 residual vaginal-rectal swabs in LIM broth collected from two clinical sites. These samples were cultured on ChromID Strepto B medium and blood agar. The digital images generated from these cultures were then analyzed by the PM CDM and manually reviewed. The results were compared against a composite reference standard derived from multiple molecular and culture-based methods.

Key Insights and Ideas

  • AI as a "second pair of eyes": The PM CDM acts as a powerful assistant, identifying subtle or overlooked GBS colonies that might be missed during manual review, even of digital images. This is highlighted by the detection of 8 additional true-positive GBS cases.
  • Bridging the sensitivity gap: The combination of ChromID Strepto B medium and the PM CDM achieved a sensitivity (95.5%) that is remarkably close to that of NAAT (96.8%), offering a more sensitive culture-based alternative.
  • Workflow optimization: By accurately identifying negative cultures, the PM CDM allows technologists to quickly batch-release a large number of negative results, significantly reducing hands-on time and allowing them to focus on potentially positive samples.
  • Specificity trade-offs: While sensitivity is greatly enhanced, the specificity of the combined ChromID + PM CDM approach (63.0%) was lower than manual review (95.4%) or NAAT (97.7%). This suggests that while the AI is excellent at finding positives, further refinement might be needed to reduce false positives.

Key Results

  • Sensitivity:
    • ChromID alone (48h): 90.3%
    • ChromID + PM CDM (48h): 95.5%
    • NAAT: 96.8%
  • Specificity:
    • ChromID alone (48h): 95.4%
    • ChromID + PM CDM (48h): 63.0%
    • NAAT: 97.7%
  • The PM CDM achieved 100% sensitivity at 48 hours of incubation when compared to manual reading of digital images, identifying all presumptive GBS colonies.
  • The algorithm never called a culture negative that was determined to be positive by manual reading.
  • An additional 8 true-positive GBS specimens were identified by the PM CDM that were missed by manual digital image culture reading.

Limitations and Future Work

  • Specificity: The study notes that the specificity of the ChromID plus PM CDM was lower (63.0%) compared to manual review or NAAT. This is a key area for improvement.
  • Algorithm Tuning: The authors suggest that as more chromogenic media are developed with improved color differentiation, algorithms can be continually modified to enhance specificity.
  • Cost of Investment: While the paper argues for the cost-effectiveness of this approach compared to NAAT in the long run, it acknowledges that a significant initial investment in technology (like the WASPLab and PM software) is necessary.
  • Future Directions: The paper implies that further development and integration of such AI-driven image analysis into routine laboratory workflows can increase the adoption of more sensitive, less expensive culture-based methods, potentially replacing NAAT in certain scenarios where rapid turnaround time is not critical.

Summary updated

Abstract

Group B <i>Streptococcus</i> (GBS) can be found to colonize about 25% of all healthy, adult women and is the leading infectious cause of early neonatal morbidity and mortality in the United States. This study evaluated the clinical performance of PhenoMatrix (PM) chromogenic detection module (CDM) digital imaging software in detection of GBS from LIM broth plated on ChromID Strepto B chromogenic medium (ChromID) using the WASP automated processor. The performance of the PM CDM was compared to manual culture review of the digital images and molecular detection of GBS. ChromID alone had a sensitivity and specificity of 84.5% and 94.7%, respectively, after 48 h compared to nucleic acid amplification testing (NAAT). Compared to the composite reference for positivity, when PM CDM was used to detect GBS from ChromID, the sensitivity was 100%, with no true-positive GBS isolates missed by 48 h of incubation. Overall, evaluating all three methods for the detection of GBS, the sensitivities of NAAT, ChromID plus PM CDM at 48 h, and ChromID alone at 48 h were 96.8%, 95.5%, and 90.3%, respectively. The specificities of NAAT, ChromID plus PM CDM, and ChromID alone were 97.7%, 63.0%, and 95.4%, respectively. The sensitivity of ChromID in combination with the PM CDM was similar to the sensitivity of molecular detection. Further, the algorithm never called a culture negative that was determined to be positive by manual reading, and it identified an additional eight true positive specimens that were missed by manual digital image culture reading.
BibTeX
@article{baker2020digital,
  author = {Justin Baker and Karen Timm and Matthew L. Faron and Nathan A. Ledeboer and Karissa Culbreath},
  title = {{Digital Image Analysis for the Detection of Group B <i>Streptococcus</i> from ChromID Strepto B Medium Using PhenoMatrix Algorithms}},
  journal = {Journal of Clinical Microbiology},
  year = {2020},
  doi = {10.1128/jcm.01902-19},
}
Figure