Introduction:

Clustering is widely used in histopathology to analyse tissue architecture, but evaluating its performance remains challenging, especially in heterogeneous samples. Rapid latent space assessment is crucial for selecting foundation models in machine learning. Therefore, we introduce a Normalized Weighted Composite Score (WCS) to quantitatively assess clustering effectiveness in histopathology patches.​

​

Methods:

Our approach combines multiple criteria—spatial coherence, morphological consistency, and cluster compactness—into a single composite score. We apply this framework to breast tissue histopathology images. To validate our method, we conducted a qualitative visual assessment using our DimReduce software [1]. By benchmarking various clustering algorithms, we demonstrate that our scoring equation offers a robust and interpretable measure of clustering quality.

​

Results:

Using WCS, we achieved a 33.65% improvement in selecting the optimal feature extractor across 20 combinations of foundation models and clustering methods on the target datasets [2,3]. By analyzing latent space embeddings, we identified the model with the highest WCS value, ensuring compact clusters with preserved morphological consistency. WCS allows adaptive weighting to accommodate different dataset characteristics. For example, increasing morphological weighting by 10% alters cluster rankings, demonstrating its ability to prioritize biologically relevant features. This flexibility makes WCS a more comprehensive evaluation metric than traditional scores, which assess only isolated aspects of clustering quality.

​

Conclusion:

WCS enhances objective clustering comparisons, improving the reliability of automated tissue analysis in digital pathology. It accelerates foundation model selection by identifying the most suitable latent space for a downstream dataset.

​

References:

[1] Wilkie, P. et al. (2024). Histology Hide-and-Seek: Navigating Latent Space Clustering for Pathology Exploration. Digital Pathology Association, Orlando, FL.
[2] Rakovitch, E., Nofech-Mozes, S. et al. (2015). Validation of the DCIS Score for Recurrence Risk Prediction. Breast Cancer Res Treat, 152(2), 389–398. 
[3] Martel, A. L., Nofech-Mozes, S., Salama, S., Akbar, S., & Peikari, M. (2019). Assessment of Residual Breast Cancer Cellularity after Neoadjuvant Chemotherapy using Digital Pathology [Data set]. The Cancer Imaging Archive. https://doi.org/10.7937/TCIA.2019.4YIBTJNO