Yaz-PaacovNurit Yaz-Paacov, PhD

Chief Scientific Officer






Image-based detection of actionable biomarkers using machine learning algorithms



Background: Orchestrating cancer treatment is a challenging task encapsulating genomic and clinical complexity. Despite significant advances in precision medicine and cancer research, much more needs to be revealed than what is understood. Digital genomics is a leading discipline in the cancer realm with great potential to tackle cancer diagnosis and treatment response challenges. During the current routine diagnostic workflow, pathologists review Hematoxylin and Eosin (H&E) stained slides as a preliminary analysis. Using digital genomics, the same slide can be utilized to directly detect biomarkers' status. Aim To demonstrate that image-based molecular diagnosis using machine-learning (ML) algorithms can serve as an accurate method for direct and fast biomarker detection.


Methods: H&E Whole slide images (WSIs) from internal and publicly available databases were used to generate different ML-based classifiers for actionable biomarkers. The data was divided into training and validation sets. Algorithm development using advanced Convolutional Neural Network (CNN) analysis making use of both supervised and unsupervised learning methods was employed. Models were validated using both retrospective and prospective cohorts.


Results: In this work, we demonstrate high performances of AI-classifiers developed to identify tens of actionable cancer biomarkers from H&E images alone. Models were created for biomarkers of seven different cancer tissues, including lung, thyroid, breast, bladder, ovary, brain, and hematological malignancies. Validation was performed on over 2,000 patients from 35 different medical centers. The average sensitivity of the various models was 94.8%, and specificity of over 90%, with an average AUC of over 0.95. These classifiers identify different types of alterations, including mutations (i.e., EGFR, BRAF), structural variants (i.e., NTRK, RET, ALK, ROS1), differential gene expression (i.e., HER2, ER, PR), and cancer signatures (i.e., HRD). Model generalization was tested by training models on samples from one medical center and inferring the status of samples from the same center as well as from a second medical center, both showing similar performance accuracies.


Conclusion: To overcome the challenges encapsulated in complex diseases, such as cancer, novel technologies need to be adopted. Artificial Intelligence (AI) is emerging as a potential tool rendering such a task feasible. The work described here presents support that the implementation of such solutions to the routine practice has the potential to dramatically improve patient care. Image based-machine learning models generate a fast, accurate, and standardized method to test all patients systematically. This process can assist the pathology workflow, clinical trial enrollment, and support treatment decisions. 



  1. At the end of this program participants should be able to understand the potential use of AI within the cancer diagnostic workflow; how AI can be used to identify cancer biomarkers and improve patient care.



Dr. Nurit Paz-Yaacov is the Chief Scientific Officer at Imagene AI - a BioMed startup company with a ground breaking technology in the field of precision oncology using Artifical Intelligence, Dr. Paz-Yaacov brings more than 20 years of multidiciplinary scientific research in human genetics and genomic sequencing and interpretation, with emphasis on cancer research. Prior to her role at Imagene AI, Dr. Paz-Yaacov was the Head of Scientific Research at Genoox, a cloud platform that provides actionable interpretation for real-time/real-life genomic data.Dr. Paz-Yaacov holds B.Sc, M.Sc, and Ph.D. degrees from Tel-Aviv University and completed her Postdoctoral studies at Bar-Ilan University, Her studies and research in the fields of cancer, applied genomics and epi-genomics, received recognition through numerous publications and prizes.

40+ Expert Presenters

View the PV22 agenda.


Register for #PathVisions22! 

October 16-18 | Las Vegas, NV


Support & Exhibit Opportunities

Trade Show Sold Out!