Histopathological image analysis for centroblasts classification through dimensionality reduction approaches.

Area of Science:

• Digital pathology
• Computational biology
• Hematopathology

Background:

• Follicular lymphoma diagnosis relies on identifying centroblast (CB) cells.
• Distinguishing CB cells from similar-looking non-CB cells in H&E-stained tissues is challenging.
• Automated image analysis can improve diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate novel automated image analysis methods for differentiating CB from non-CB cells.
• To compare the performance of these methods against existing classification techniques.
• To assess the reproducibility of the proposed classification methods.

Main Methods:

• Two novel methods were developed: one using Singular Value Decomposition (SVD) and another using Laplacian Eigenmaps.
• These methods approximate intrinsic dimensionality to extract discriminatory features from cell image subspaces.
• A database of 213 CB and 234 non-CB regions of interest from 500 high-power field images was utilized for validation.

Main Results:

• Both methods demonstrated high classification accuracy, with average overall accuracy rates of 99.22% ± 0.75% for the SVD-based method and 99.07% ± 1.53% for the Laplacian Eigenmaps method.
• The developed methods outperformed existing classification techniques in accuracy.
• Reproducibility studies confirmed the reliability of both classification approaches.

Conclusions:

• The proposed automated image analysis methods effectively differentiate centroblast and non-centroblast cells in follicular lymphoma.
• These novel techniques offer a significant improvement over current state-of-the-art methods for CB/non-CB classification.
• The high accuracy and reproducibility suggest potential for clinical application in lymphoma diagnostics.