Development and Evaluation of Artificial Intelligence-Based Two-Step Model for Automated Serum Quality Assessment in...

Area of Science:

• Clinical Chemistry
• Artificial Intelligence in Medicine
• Medical Diagnostics

Background:

• Accurate pre-analytical quality control (QC) is vital in clinical laboratories.
• Identifying sample interferences like hemolysis, icterus, or lipemia is crucial for reliable test results.
• Automated QC can mitigate risks associated with subjective manual assessments.

Purpose of the Study:

• To evaluate a novel, self-developed artificial intelligence (AI) approach for assessing serum quality.
• To explore the performance of a deep learning model (YOLOv5-ResNet-50) under real-world laboratory conditions.
• To enhance pre-analytical QC and provide risk alerts for common sample interferences.

Main Methods:

• A two-step deep learning architecture (YOLOv5-ResNet-50) was developed for serum quality assessment.
• The AI model was integrated into a Roche Cobas c 701 module for automated preprocessing.
• A laboratory-specific model was created using a dataset of 42,000 serum samples (21,000 training/validation, 21,000 independent testing).

Main Results:

• The AI serum model achieved >97% accuracy for hemolysis and icterus detection.
• Recognition accuracy for lipemia exceeded 92%.
• The AI model showed higher consistency (72% exact agreement) compared to manual assessments (42%-68% exact agreement).

Conclusions:

• AI-powered serum quality software demonstrates high accuracy in pre-analytical assessment.
• The AI tool meets or exceeds the performance of automated and manual measurements.
• This technology enhances laboratory efficiency and reduces error risks from subjective visual evaluations.