CELLAVISION DM1200 WITH THE BODY FLUID APPLICTION

The CeliaVision® DM1200 with the body fluid application is an automated system for in vitro diagnostic differential counting of white blood cells in cytocentrifuged body fluid preparations. It utilizes a 'battlement track pattern' for cell location and artificial neural networks for image-based pre-classification, which is then verified by skilled operators, showing strong accuracy and precision comparable to established methods.

The CeliaVision® DM1200 with the body fluid application is an automated in vitro diagnostic system designed for performing differential counts of white blood cells (WBCs) from various body fluids, including cerebrospinal, serous, and synovial fluids. The system processes cytocentrifuged and Romanowsky-stained slide preparations by automatically locating nucleated cells using a "battlement track pattern" and capturing their images. These images are then analyzed using standard mathematical methods and deterministic artificial neural networks (ANNs) trained to pre-classify WBCs into categories such as neutrophils, eosinophils, lymphocytes, macrophages, and an "Other" category, which includes rarer or abnormal cells like basophils, lymphoma cells, blasts, and tumor cells. Skilled operators subsequently review these pre-classifications, verifying or reclassifying cells as needed, and can view an overview of the entire slide at different magnifications. Performance evaluation of the DM1200 involved a method comparison study against the predicate CeliaVision DM96, using 260 body fluid samples. For each sample, a 400-cell differential count (200 cells per slide, two slides per sample) was performed, with results verified by skilled human operators. Accuracy results, presented as regression analyses, showed high correlation between the DM1200 and the reference method for major cell classes, with R² values ranging from 0.9273 for "Other cells" to 0.9932 for neutrophils. For instance, neutrophil classification yielded a slope of 0.9969 and R² of 0.9932, while lymphocyte classification showed a slope of 0.9815 and R² of 0.9829. Short-term imprecision was assessed and found to be equivalent between the DM1200 and the predicate device, demonstrating comparable mean percentages and standard deviations for identified cell types.

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 system's 'battlement track pattern' for cell locating compare in efficiency and slide coverage to other automated microscopy scanning methodologies?; What specific morphological features or image parameters are most heavily weighted by the artificial neural networks when distinguishing between challenging cell classifications, particularly for cells initially grouped into the 'Other' category?; Are there notable performance differences or specific calibration considerations when the system analyzes various body fluid types (e.g., CSF vs. synovial fluid) due to differences in typical cellularity, viscosity, or background matrix?; What is the typical processing time per slide for a standard 200-cell differential count, encompassing both automated analysis and the necessary operator verification steps?; Given the R² value of 0.9273 for 'Other cells' is lower than for more defined cell types, what are the specific cell types within this category that prove most challenging for the ANN, and are there ongoing efforts to improve their pre-classification accuracy?