Smarter Digital Morphology Triage - Summary - MDSpire

Smarter Digital Morphology Triage

A diffusion-based approach aims to handle domain shifts, slide variability, and rare cell appearances

  • January 30, 2026

  • 3 min

Share

Objective:

To evaluate the performance of CytoDiffusion, a generative AI model for classifying blood cell images and identifying morphologies requiring specialist review.

Key Findings:
  • CytoDiffusion's synthetic morphology was difficult to distinguish from real images, indicating high realism.
  • The model achieved a sensitivity of 0.91 and specificity of 0.96 in detecting abnormal cells.
  • CytoDiffusion outperformed traditional models in handling uncertainty and anomaly detection.
Interpretation:

The findings suggest a shift towards generative models in digital morphology that can quantify uncertainty and recognize rare cell patterns, enhancing diagnostic workflows.

Limitations:
  • CytoDiffusion is more computationally intensive than conventional classifiers, with a mean classification time of approximately 1.8 seconds per image.
Conclusion:

CytoDiffusion represents a promising advancement in digital morphology, potentially improving triage processes in hematology diagnostics.

Original Source(s)

Smarter Digital Morphology Triage

  • January 30, 2026

Related Content

Research Leaders Across New York State Form Empire State Cellular Therapy Consortium

To expand on early progress in the emerging field of cell and gene therapy, five research centers across New York State are collaborating to develop new cell therapies for cancer and other diseases.

  • by

  • July 29, 2025

  • 7 min
Evaluating the Impact of Pictorial versus Teach-Back Educational Approaches on Self-Efficacy and Quality of Life in Leukemia Patients: A Randomized Controlled Study

Impact of haematopoietic stem cell transplantation for benign and malignant haematologic and non-haematologic disorders on fertility: a systematic review and meta-analysis