Data Alchemy: Mitigating Cross-Site Model Variability Through Test Time Data Calibration

TL;DR

This work tackles cross-site variability in histopathology imaging and regulatory barriers to site-specific model updates. It introduces Data Alchemy, an explainable stain normalization method coupled with test-time data calibration that learns a site-alignment template without altering model weights. Empirical results on CAMELYON16 demonstrate improved stain fidelity (lower $cycleL1$, higher $SSIM$/$PSNR$ and $AP(i,p)$) and substantial cross-site gains in tumor-classification performance (AUPR up to $0.852$, often nearing or surpassing an upper-bound model). By enabling rapid, regulatory-friendly deployment of pre-trained clinical tools across sites, Data Alchemy reduces domain gaps with minimal operational overhead and enhances generalizability in digital pathology.

Abstract

Deploying deep learning-based imaging tools across various clinical sites poses significant challenges due to inherent domain shifts and regulatory hurdles associated with site-specific fine-tuning. For histopathology, stain normalization techniques can mitigate discrepancies, but they often fall short of eliminating inter-site variations. Therefore, we present Data Alchemy, an explainable stain normalization method combined with test time data calibration via a template learning framework to overcome barriers in cross-site analysis. Data Alchemy handles shifts inherent to multi-site data and minimizes them without needing to change the weights of the normalization or classifier networks. Our approach extends to unseen sites in various clinical settings where data domain discrepancies are unknown. Extensive experiments highlight the efficacy of our framework in tumor classification in hematoxylin and eosin-stained patches. Our explainable normalization method boosts classification tasks' area under the precision-recall curve(AUPR) by 0.165, 0.545 to 0.710. Additionally, Data Alchemy further reduces the multisite classification domain gap, by improving the 0.710 AUPR an additional 0.142, elevating classification performance further to 0.852, from 0.545. Our Data Alchemy framework can popularize precision medicine with minimal operational overhead by allowing for the seamless integration of pre-trained deep learning-based clinical tools across multiple sites.

Figures (7)

• Figure 1: Stain normalization during training (left) and inference (right).
• Figure 2: Data Alchemy uses the normalization network and a classifier to learn a template at test time to improve the classifier performance when deployed at a site.
• Figure 3: Qualitative visualization of stain normalization. A detailed version is available in appendix \ref{['app:stain_norm-comp']}.
• Figure 4: Eigenvalue and vectors of a content, stain, and stain normalized patch projected onto a 3D sphere. Each point of the surface represents an eigenvector and the color represents an eigenvalue. Red signifies a smaller and blue higher eigenvalues.
• Figure 5: Eigenvalue blending of the content and stain patches produces different staining. The (%) in the arrows are the content and stain patches blending weights.