CytoDiffusion is a generative AI model designed to classify blood cell images and identify morphologies needing specialist review.
2
Unlike traditional discriminative models, CytoDiffusion learns the visual range of blood cell classes, improving performance in varied real-world conditions.
3
The model was trained on 32,619 blood cell images and demonstrated high agreement with expert hematologists in classifying synthetic cells.
4
CytoDiffusion outperformed established classifiers in abnormal cell detection, achieving sensitivity of 0.91 and specificity of 0.96.
5
The model's computational intensity poses a challenge for high-throughput implementation, with an average classification time of 1.8 seconds per image.
