CellaVision DC-1, CellaVision DC-1 PPA

The CellaVision DC-1 is an automated cell-locating device that utilizes digital microscopy and Convolutional Neural Network-based artificial intelligence to analyze peripheral blood smears for white blood cell differentials, red blood cell morphology, and platelet estimation. Its methodology involves automated scanning of stained slides, image capture, software-assisted pre-classification of cells, and final verification by a skilled operator, aiming to provide accurate and reproducible hematological data.

The CellaVision DC-1 is an automated cell-locating device designed for in-vitro diagnostic use in clinical laboratories to assist trained operators in performing differential counts of white blood cells (WBC), characterizing red blood cell (RBC) morphology, and estimating platelet (PLT) levels from peripheral blood smears. The system processes Romanowsky-stained blood smears on glass slides, typically from samples flagged for abnormal morphology by automated cell counters. It employs a high-power magnification microscope with LED illumination, an XY stage, and a proprietary camera to automatically locate blood cells in the monolayer of a smear, operating on one slide at a time. The CellaVision DM Software, utilizing Convolutional Neural Networks (CNN) for artificial intelligence, pre-classifies located WBCs and pre-characterizes RBC morphology, presenting these images and preliminary classifications for review and verification by a skilled operator, who also estimates PLT levels from an overview image. Clinical performance evaluations demonstrated the CellaVision DC-1's capabilities relative to its predicate, the CellaVision DM1200, which uses Artificial Neural Networks (ANN). Repeatability and reproducibility studies for WBC, RBC, and PLT parameters conducted across multiple clinical sites were generally successful, meeting predefined acceptance criteria. Method comparison studies involving hundreds of samples showed strong agreement and accuracy between the DC-1 and DM1200 for WBC differentials (overall agreement ~89-91%), RBC morphology characterization (overall agreement ~80-88% for color, size, shape), and PLT estimation (Cohen's Kappa coefficient of 0.894). Key differences from the predicate include the DC-1's use of CNN, a single-slide capacity, manual immersion oil application, and a more integrated physical design; these differences were not found to raise new questions of safety or effectiveness.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the performance of the Convolutional Neural Network (CNN) in the CellaVision DC-1 for WBC pre-classification compare to the Artificial Neural Network (ANN) in the DM1200, particularly for rare or atypical cell morphologies not extensively covered in the provided comparison data?; What is the impact of manual immersion oil application and single-slide capacity in the DC-1 versus automatic application and multi-slide capacity in the DM1200 on workflow efficiency and inter-operator variability in various laboratory settings?; Given the overall agreement for RBC color characterization was approximately 80%, what specific RBC color variations contribute most to an operator needing to modify pre-characterizations, and are there plans to refine the algorithm for these specific variations?; What were the specific acceptance criteria for the repeatability and reproducibility studies (e.g., for variance components or agreement percentages), and for the few instances where criteria were not fully met, what were the root causes and their potential clinical significance?; How does the system differentiate between K2EDTA and K3EDTA anticoagulated samples in practice, and were any performance differences noted during validation despite literature suggesting minimal clinical significance?