Improving Turnaround Times with Artificial Intelligence in Microbiology

This dual-center study demonstrates that integrating AI-based software (PhenoMATRIX) into fully automated microbiology laboratories significantly reduces urine culture turnaround times in both low-volume and high-volume settings, with additional gains achieved through automated result release.

The Gist

Imagine a busy hospital kitchen where chefs (the lab technicians) are trying to prepare thousands of meals (urine culture tests) every day. The goal is to get the food to the customers (patients and doctors) as fast as possible so they know if they need medicine.

For a long time, these chefs had to stand over the ovens, constantly checking every single tray of food to see if it was ready. If a tray looked empty or just had a few harmless crumbs, they still had to walk over, look at it, write down a note, and stamp it "Ready to Serve." This took a lot of time and energy, especially when the kitchen was crowded and the chefs were tired.

The Problem:
The kitchen was getting busier, but there weren't enough chefs. The "meals" (urine samples) were piling up, and it was taking too long to tell the customers if they had an infection or if they were just fine.

The Solution: The "Smart Camera" (Artificial Intelligence)
The researchers in this paper installed a new system called PhenoMATRIX. Think of this as a super-smart, tireless robot camera that watches the ovens 24/7.

Instead of a human chef having to walk over and squint at every tray, this robot takes a high-definition photo of the food after 16 hours. Its "brain" (Artificial Intelligence) instantly analyzes the photo and says:

• "This tray is empty. No bacteria grew. It's safe."
• "This tray has a tiny bit of harmless dust. Ignore it."
• "This tray has a big, scary monster (bad bacteria). Alert the chef immediately!"

The Results: Two Different Kitchens
The study tested this robot in two very different kitchens:

1. The Small, Specialized Kitchen (QEII Hospital): This place handles complex, tricky cases.

   • Before: It took a long time to get results because humans had to check everything manually.
   • After: The robot did the heavy lifting. It automatically stamped the "safe" trays and sent the results to the patient's chart instantly.
   • The Win: They saved about 1.5 hours per test. That's like getting your lunch order 90 minutes faster!

2. The Massive, 24/7 Diner (Dynacare Lab): This place is huge, processing a mountain of samples every day.

   • Before: The chefs were overwhelmed, and the line for results was very long.
   • After: The robot sorted the "safe" trays from the "dangerous" ones instantly.
   • The Win: They saved nearly 4 hours per test! That's a massive jump, meaning patients got their answers almost a full work-shift earlier.

The "Auto-Release" Feature (PM+)
At the smaller hospital, they added a special feature called PM+. Imagine a conveyor belt that automatically drops the "safe" meals right into the customer's hands without the chef even touching them.

• If the robot sees "No Growth," it automatically sends the report to the doctor.
• This freed up the human chefs to focus only on the tricky, dangerous trays that actually needed their expert eyes.

Why This Matters
In the real world, doctors are waiting for these results to decide if a patient needs antibiotics. Every hour saved is an hour a patient spends in less pain or an hour a doctor can spend on a different patient.

The Bottom Line:
This paper shows that by giving our lab "chefs" a smart robot assistant that can read the food trays, we can speed up the whole process. It doesn't replace the chefs; it just lets them do their most important work faster, while the robot handles the boring, repetitive checking. The result? Faster answers for patients and a happier, less stressed kitchen.

Technical Summary

1. Problem Statement

Clinical microbiology laboratories face persistent challenges due to staffing shortages and increasing specimen volumes. While full laboratory automation (e.g., WASPLab systems) has improved efficiency over manual methods, there remains a need to further optimize workflows to reduce Turnaround Time (TTRR) for culture results. Specifically, the manual review of culture plates by technologists creates bottlenecks, particularly for high-volume urine cultures where a significant portion of results are negative or non-significant. The study addresses the question of whether Artificial Intelligence (AI) can further accelerate result reporting by automating plate assessment and result release.

2. Methodology

The study was a dual-center evaluation conducted in Canada between May 2022 and May 2025, comparing performance before and after the implementation of Copan's PhenoMATRIX (PM) software.

• Study Sites:
  • Site A (QEII): A low-to-medium volume tertiary care hospital in Halifax, Nova Scotia (~65,000 urine specimens/year).
  • Site B (Dynacare): A high-volume community laboratory in Brampton, Ontario (~650,000 urine specimens/year).
• Intervention:
  • PhenoMATRIX (PM): AI software that digitally images culture plates after 16 hours of incubation. It uses algorithms to categorize growth (no growth, <10 colonies, mixed growth, specific chromogenic colors) and sort plates.
  • PhenoMATRIX PLUS (PM+): An advanced feature implemented at QEII that automatically releases defined results (negative or normal flora) to the Laboratory Information System (LIS) without human intervention.
• Data Collection:
  • Metrics: Time to Result Reporting (TTRR) was measured from specimen placement on the automation line (or receipt at QEII) to final report release in the LIS.
  • Phases: Data was collected during Pre-implementation (PrePM), Post-implementation of PM (PostPM), and Post-implementation of PM+ (PostPM+ at QEII only).
  • Controls: No other significant workflow or staffing changes occurred during the study periods to ensure data validity.

3. Key Contributions

• First Dual-Site North American Evaluation: This is the first study to evaluate the impact of PM AI algorithms on urine culture TTRR across two distinct laboratory environments (tertiary hospital vs. high-volume community lab).
• Validation of AI in Diverse Settings: The study demonstrates that AI-driven plate interpretation improves efficiency regardless of laboratory size, specimen volume, or baseline workflow differences.
• Proof of Concept for Auto-Release: The study validates the efficacy of PM+, showing that fully automating the release of negative/normal results significantly reduces the time to final reporting.

4. Key Results

The implementation of AI resulted in significant reductions in TTRR at both sites:

A. QEII (Tertiary Care Hospital)

• Volume: Increased ~7% during the study period.
• TTRR Reduction: Mean TTRR decreased by approximately 1 hour and 26 minutes (approx. 1.5 hours) between June 2023 and May 2025.
• Impact of PM+ (Auto-release):
  • 17 Hours: Final reports released increased from 8.6% (PrePM) to 26.3% (PostPM+).
  • 18 Hours: Reports released increased from 34.1% (PrePM) to 51.9% (PostPM+).
  • 24 Hours: 75.6% of results were released with PM+, compared to 69.0% previously.
  • Pending Results: The proportion of cultures pending beyond 24 hours dropped from 31.0% to 24.4%.

B. Dynacare (High-Volume Community Lab)

• Volume: Increased ~11.6% during the study period.
• TTRR Reduction: Mean TTRR improved by approximately 3 hours and 51 minutes (approx. 3.9 hours) from May 2022 to May 2023.
• Impact of PM:
  • 17 Hours: Reports released increased from 10.5% (PrePM) to 25.9% (PostPM).
  • 18 Hours: Reports released increased from 20.8% (PrePM) to 47.9% (PostPM).
  • 24 Hours: Reports released increased from 56.7% (PrePM) to 72.4% (PostPM).
  • Pending Results: The proportion of cultures pending beyond 24 hours decreased by 15.7% (from 43.3% to 27.6%).

5. Significance and Conclusion

• Operational Efficiency: The study confirms that integrating AI into fully automated microbiology workflows allows for continuous plate assessment, enabling earlier result release and reducing the backlog of pending cultures.
• Resource Optimization: By automating the review and release of negative or non-significant cultures (which constitute ~50–77% of urine cultures), laboratories can redirect skilled technologists toward complex, positive, or mixed-growth cases.
• Future Outlook: The authors suggest that combining smart incubation, automated imaging, and AI interpretation represents a major advancement toward end-to-end automation. This evolution promises to maintain high throughput and timely patient results despite workforce shortages, potentially allowing for fully automated identification and susceptibility testing for uncomplicated cultures in the future.

In conclusion, the integration of AI-based culture assessment and auto-release features offers a measurable, scalable solution to improve turnaround times in modern diagnostic microbiology.