RudmannDaniel G. Rudmann, DVM, PhD, DACVP, FIATP

Scientific Director

Charles River Laboratories




Advancing Non-Clinical Pathology Using Decision Support Deep Learning


Background: Toxicologic veterinary pathologists examine millions of tissues on glass slides for non-clinical research supporting biotechnology, pharmaceutical, medical device, and chemical industries. Pre-analytical and analytical variability is considerable as is the diversity of the backgrounds for these highly trained scientists. It is estimated that approximately 80% of samples evaluated by the toxicologic pathologist are normal and analyses of lesioned tissues is largely qualitative. Deep learning-based artificial intelligence is positioned well to provide diagnostic decision support for the non-clinical pathologist.


Methods: Hematoxylin and eosin-stained tissues from toxicology studies were scanned at 40x magnification. Several target organs were selected for developing supervised convolutional neural network (CNN)-based deep learning models. Annotation classes were based on INHAND diagnostic criteria and training was done by board certified veterinary pathologists. Confusion matrices and F score analyses were used to guide training and a 4-tier qualification procedure was adopted for validation. Binary (lesion or no lesion) and diagnostic models were designed for screening and high sensitivity.


Results: Both binary and diagnostic CNN AI models were built for rodent kidney, liver, nasal cavity, stomach, lung, broncho-alveolar lavage (BAL), thyroid gland, adrenal gland, and testes. Qualification results supported the decision support intended use. The models presented normal and lesion (binary or diagnostic) colored tisue masks within a digital slide browser to the non-clinical pathologists for decision support during slide evaluation.


Conclusion: AI tissue decision support models demonstrated potential for supporting the non-clinical toxicologic pathologist in both toxicity screening (tissue lesion detection and diagnostics) and quantitation (BAL cytology) workflows.



  1. Define convolutional neural network (CNN)- based AI for computer vision
  2. Describe a quality system and 4-level qualification approach for non-clinical CNN-based AI models
  3. Describe the potential application of CNN-based AI in non-clinical pathology decision support



Dr. Dan Rudmann has almost 25 years of experience in the pharmaceutical and biotechnology industries and is presently Scientific Director, Digital Pathology at Charles River Laboratories. In this role, Dr. Rudmann is leading the digital transformation for pathology at Charles River. Dr. Rudmann has worked across 4 different scientific disciplines (pharmaceutical toxicology, translational medicine, veterinary and clinical development and biotechnology) as a scientist, group leader, head/director/Vice President, project manager, and Six Sigma Black belt), all unified by the broad and deep comparative medical expertise gained from DVM, anatomic pathology residency, and research (PhD) training. Dr. Rudmann has maintained several leadership roles in industry as well as in the American College of Veterinary Pathology and Society of Toxicologic Pathology and has published over 30 manuscripts.

Over 30 Expert Presenters!

View the PV21 agenda.


Register for #PathVisions21! 

October 17-19 | Las Vegas, NV


Support & Exhibit Opportunities

Limited booth space remaining!