Unsupervised Cellular Anomaly Detection in Toxicological Histopathology

Saketh Juturu, Geetank Raipuria, Raghav Amaravadi, Aman Srivastava, Malini Roy, Nitin Singhal

Keywords: Anomaly Detection, Out-of-distribution Detection, Toxicology, Histopathology, Foundation Models, Cellular Analysis, Drug Safety Assessment

TL;DR: We evaluate cellular anomaly detection for toxicological histopathology, and find that KNN-distance based method using best-performing foundation models outperforms state-of-the-art methods including DDPMs.

Abstract: Irregularities in cellular representation play a crucial role in assessing drug-induced tissue alterations in toxicological histopathology studies. However, the process of annotating rare abnormal cellular variations for training supervised deep learning models presents significant challenges and lacks scalability. While anomaly detection is well-suited for this purpose, it has not yet been explored for cellular-level analysis. In this study, we evaluate cellular anomaly detection using datasets derived from the kidney and liver tissue of Wistar rats. Our findings show that a KNN-distance-based anomaly detection method significantly benefits from employing a feature extractor that has been pre-trained on extensive unsupervised histopathology datasets. When utilizing the best-performing feature extractor, the KNN-distance method surpasses state-of-the-art anomaly detection models by over 4.84% (AUC), including the denoising diffusion probabilistic model, in detecting cellular anomalies. Additionally, we assess the effectiveness of this method in identifying variations in anomalous cell counts between control and treated animal tissues within a toxicological study, revealing a statistically significant difference between the two dosage groups.

Primary Subject Area: Unsupervised Learning and Representation Learning

Secondary Subject Area: Application: Histopathology

Paper Type: Validation or Application

Registration Requirement: Yes

Visa & Travel: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Created a single midl25_NNN.zip file with midl25_NNN.tex, midl25_NNN.bib, all necessary figures and files., Includes \documentclass{midl}, \jmlryear{2025}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package, Did not use the times package., All authors and co-authors are correctly listed with proper spelling and avoid Unicode characters., Author and institution details are de-anonymized where needed. All author names, affiliations, and paper title are correctly spelled and capitalized in the biography section., References must use the .bib file. Did not override the bibliographystyle defined in midl.cls. Did not use \begin{thebibliography} directly to insert references., Tables and figures do not overflow margins; avoid using \scalebox; used \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., Appendices and supplementary material are included in the same PDF after references., Main paper does not exceed 9 pages; acknowledgements, references, and appendix start on page 10 or later.

Submission Number: 145