Single‐Cell Morphometrics From Hematoxylin and Eosin‐Stained Images Reveals Subtype‐Specific Features in Chronic Lymphocytic Leukemia

Abstract

Chronic lymphocyticleukemia (CLL) is a highly heterogeneous B-cell malignancy, spanning a spectrumfrom indolent conventional forms (C-CLL) to Richter transformation. AcceleratedCLL (A-CLL) represents an intermediate subtype, histologically and clinicallymore aggressive than C-CLL, yet its diagnosis relies on subjective histologicalcriteria. We hypothesized that high-dimensional single-cell morphometrics fromroutine hematoxylin and eosin-stained lymph node biopsies could providecomplementary metrics for positioning cases along the disease continuum. Weimplemented a novel, open-source methodology using CellProfiler and Seurat tosegment cells and measure hundreds of parameters from two indolent and twoaggressive cases. Datasets were integrated to mitigate technical batch effects.Our analysis combined an unsupervised approach with trajectory analysis and apathologist-guided supervised machine learning model. Both unsupervised andsupervised approaches successfully and independently distinguished aggressivefrom indolent cases. Aggressive cases were enriched in morphometric clusterscorresponding to paraimmunoblasts, while indolent cases were enriched in thosecorresponding to small lymphocytes. Pseudo-bulk analysis using a subset of keyparameters also successfully classified patients. In thishighly exploratory, proof-of-concept study utilizing a small, well-characterizedcohort (n=4 CLL patients plus 2 RT reference cases), we present a robust and traceablesingle-cell morphometric pipeline. By providing objective metrics of cellularand subcellular morphology, our method reveals strong segregation patternsreflecting the CLL continuum. Our findings warrant validation in largercohorts, but offer novel, quantifiable insights that could potentiallycomplement diagnostic criteria and aid in patient stratification.