Related Experiment Video
Updated: May 9, 2025
12:05
Database-guided Flow-cytometry for Evaluation of Bone Marrow Myeloid Cell Maturation
Published on: November 3, 2018
11.5K
Comparison between five pattern-based approaches for automated diagnostic classification of mature/peripheral B-cell
C E Pedreira1, Q Lecrevisse2, R Fluxa3
1Systems and Computing Department (PESC), COPPE, Federal University of Rio de Janeiro (UFRJ), Brazil.
Computers in Biology and Medicine
|April 29, 2025
Summary
Five artificial intelligence algorithms accurately classify B-cell chronic lymphoproliferative disorders (B-CLPD) using flow cytometry data. These AI tools offer a promising approach for precise B-CLPD diagnosis, improving upon existing methods.
Area of Science:
- Hematology
- Immunology
- Computational Biology
- Artificial Intelligence in Medicine
Background:
- Flow cytometry immunophenotyping is crucial for diagnosing B-cell chronic lymphoproliferative disorders (B-CLPD).
- Quantitative classification approaches can maximize information extraction from multiparameter flow cytometry data for automated patient classification.
Purpose of the Study:
- To develop and compare five diagnostic classification algorithms for B-CLPD.
- To evaluate the accuracy, precision, and coverage of these algorithms using a large dataset of B-CLPD patient samples.
Main Methods:
- Five algorithms were developed: Principal Component Analysis (PCA), Canonical Variate Analysis (CVA), Neighbourhood Component Analysis (NCA), Support Vector Machine (SVM), and a variant of Canonical Analysis (CA-vSD).
- Algorithms were trained and tested on a multicentric EuroFlow dataset of 659 B-CLPD patients, classified according to WHO criteria.
- Performance was evaluated based on accuracy (correctly classified cases), precision (single vs. multiple diagnoses), and coverage (proposed diagnoses).
Main Results:
- Average correct diagnosis rates ranged from 58.9% to 90.6% across the five algorithms.
- PCA, SVM, and CA algorithms showed high correctness (86.0%-90.6%) but often proposed multiple diagnoses.
- CA-vSD achieved the lowest misclassification rate (4.1%) but had the highest unclassified rate (37.0%); NCA had minimal unclassified cases (2.7%) but higher misclassification (14.0%).
Conclusions:
- The developed AI algorithms provide acceptable accuracy for B-CLPD diagnostic classification.
- These algorithms generally surpass previously reported methods in classifying B-CLPD patients.
- The choice of algorithm involves balancing accuracy, precision, and coverage for optimal diagnostic utility.
