Automatic area classification in peripheral blood smears.

Area of Science:

• Computational Pathology
• Medical Image Analysis
• High-Throughput Screening

Background:

• Peripheral blood smear analysis is crucial for biological and pathological diagnoses.
• Manual selection of optimal smear areas is subjective, time-consuming, and prone to bias.
• Existing research on automated working-area classification for blood smears is limited.

Purpose of the Study:

• To develop an integrated algorithm for automatic classification of appropriate working areas on peripheral blood smears.
• To provide a preprocessing step for detailed cell enumeration and diagnosis in high-throughput screening (HTS).
• To quantify cell spreading and clumping characteristics within smear images.

Main Methods:

• An integrated algorithm was developed for automatic area classification of blood smear images.
• The algorithm quantifies cell spreading and clumping, including individual clumps and group occurrence probabilities.
• Experiments were conducted using Giemsa-stained peripheral blood smear images.

Main Results:

• The proposed method achieved high accuracy, with hit rates above 88.9% on a validation set (140 images).
• Robustness was demonstrated with hit rates exceeding 78.1% on a large test set (4878 images).
• Comprehensive comparisons validated the effectiveness of cell spreading and clumping quantifications.

Conclusions:

• The developed algorithm efficiently and accurately identifies suitable working areas on blood smears.
• This automated approach significantly improves upon manual selection methods in terms of reproducibility and statistical bias.
• The method provides a strong foundation for rapid working-area selection in high-throughput screening applications.