HistoSmith: Single-Stage Histology Image-Label Generation via Conditional Latent Diffusion for Enhanced Cell Segmentation and Classification

TL;DR

Histology labeling datasets are scarce, hindering robust cell segmentation and classification. HistoSmith uses a single-stage latent diffusion model that learns a joint distribution of cellular layouts, semantic masks, and tissue images in a latent space learned by a VQ-VAE, conditioned on a 10-D vector of tissue type and cell counts to generate image-label pairs. On CoNIC and CytoDArk0, augmenting the training data with HistoSmith-produced samples yields average CISCA gains of 1.9% and 3.4%, with pronounced improvements for neutrophils and brain regions such as the hippocampus and visual cortex, illustrating improved generalization. The work demonstrates controllable, data-efficient augmentation for histology analysis and highlights considerations about distribution drift and computational cost.

Abstract

Precise segmentation and classification of cell instances are vital for analyzing the tissue microenvironment in histology images, supporting medical diagnosis, prognosis, treatment planning, and studies of brain cytoarchitecture. However, the creation of high-quality annotated datasets for training remains a major challenge. This study introduces a novel single-stage approach (HistoSmith) for generating image-label pairs to augment histology datasets. Unlike state-of-the-art methods that utilize diffusion models with separate components for label and image generation, our approach employs a latent diffusion model to learn the joint distribution of cellular layouts, classification masks, and histology images. This model enables tailored data generation by conditioning on user-defined parameters such as cell types, quantities, and tissue types. Trained on the Conic H&E histopathology dataset and the Nissl-stained CytoDArk0 dataset, the model generates realistic and diverse labeled samples. Experimental results demonstrate improvements in cell instance segmentation and classification, particularly for underrepresented cell types like neutrophils in the Conic dataset. These findings underscore the potential of our approach to address data scarcity challenges.

Figures (6)

• Figure 1: Ten HistoSmith-generated samples of the colon (rows 1-2), hippocampus (row 3), cerebellum (row 4), and auditory cortex (row 5), each with a generated image, distance map, cell type semantic mask, and post-processed label overlay.
• Figure 2: The proposed HistoSmith framework for generative data augmentation.
• Figure 2: Assessment of the quality of generated vs. test images.
• Figure 3: Cell distribution per image across tissue/cell types in datasets $\mathcal{D}$ and $\hat{\mathcal{D}}$.
• Figure 4: Correlation between conditioning ($n_i$) and generated cell quantities ($n_o$).