Distinguishing Reactive Lymphocytes From Blasts Using Fractal Chromatin Patterns

Area of Science:

• Hematology
• Computational Biology
• Image Analysis

Background:

• Reactive lymphocytes (RLs) and blasts are morphologically similar and difficult to differentiate in peripheral blood smears.
• Distinguishing RLs from blasts is critical as they are associated with different diseases, prognoses, and treatments.
• Current morphological methods lack definitive accuracy in distinguishing these cell types.

Purpose of the Study:

• To develop and validate a computational method for differentiating reactive lymphocytes (RLs) and blasts.
• To assess the efficacy of fractal chromatin pattern quantification in distinguishing between RLs and blasts.
• To explore the potential clinical utility of this algorithm in hematology laboratories.

Main Methods:

• Nuclei from white blood cell images were isolated for fractal pattern quantification using TWOMBLI software.
• Machine learning models (random forest, k-nearest neighbors) were trained and tested using cross-validation.
• Performance metrics including accuracy, AUC, precision, specificity, and sensitivity were evaluated.

Main Results:

• The developed classification algorithm achieved an average accuracy of 84.2% and an AUC of 0.844 in distinguishing RLs and blast subtypes.
• All models demonstrated an area under the curve (AUC) greater than 0.815 on the testing set.
• Principal Component Analysis (PCA) identified two key components explaining 50% of the data variance.

Conclusions:

• Fractal chromatin pattern analysis provides an effective means to distinguish between reactive lymphocytes and blasts.
• The proposed classification algorithm shows promise for assisting clinicians in differentiating these critical cell types in peripheral blood smears.
• This approach may enhance diagnostic accuracy and improve patient management in clinical hematology.