Computer-aided detection of centroblasts for follicular lymphoma grading using adaptive likelihood-based cell...

Area of Science:

• Pathology
• Computational Biology
• Medical Imaging

Background:

• Follicular lymphoma (FL) is a common lymphoid malignancy with variable clinical outcomes.
• Histological grading of FL relies on manual counting of centroblasts (CB), a process that is subjective and prone to inter/intra-reader variability.
• Accurate CB identification is crucial for treatment decisions in FL patients.

Purpose of the Study:

• To develop and evaluate a computer-aided detection (CAD) system for automated identification of centroblasts (CB) in H&E-stained follicular lymphoma (FL) tissue samples.
• To address the limitations of manual histological grading, including subjectivity and reader variability.

Main Methods:

• A unitone conversion technique was employed to generate a single-channel image with maximal contrast from H&E-stained FL tissue sections.
• A cell-likelihood image was created from the contrast-enhanced image, leveraging the bimodal distribution characteristic of H&E staining.
• A two-step CB detection algorithm was implemented: first, non-CB cells were identified based on size and shape; second, CB detection was refined using texture analysis of non-CB cells.

Main Results:

• The proposed CAD system achieved a promising detection accuracy of 80.7% on 100 region-of-interest images from ten distinct FL tissue samples.
• The automated approach demonstrated potential in overcoming the subjectivity and variability associated with manual CB counting.

Conclusions:

• The developed computer-aided detection system offers a viable solution for automated identification of centroblasts in follicular lymphoma.
• This automated method has the potential to improve the consistency and objectivity of FL histological grading, aiding in clinical treatment decisions.