TEL AVIV, Israel, August 30, 2022 (Newswire.com) - Imagene, an emerging leader in the field of AI-based precision oncology, today announced a collaborative research with the pathology department at the Tel Aviv Sourasky Medical Center. The research "Image-based identification of HER2 status in H&E-stained breast cancer slides" was selected to be presented at the European Congress of Pathology (ECP), Sept. 3-7, 2022, Basel, Switzerland.
Breast cancer treatment is based, among other, on the Human Epidermal growth factor Receptor 2 (HER2) status. Current methodologies for HER2 testing rely on immunohistochemistry (IHC) followed by analysis using in situ hybridization (ISH) techniques in equivocal (score 2) cases. The interpretation of those conventional methodologies is limited due to the subjective nature of the method. Thus, major differences are seen between pathologist analysis or different medical centers since IHC scoring is based on a signal quantification that, in most cases, is done manually. Discrepancies can be as broad as 50% in close categories (score 1 vs. 2, score 2 vs. 3, etc.).
Imagene developed a deep learning algorithm to directly infer HER2 status in real-time from an H&E-stained slide image alone. In this research, the HER2-classifier was tested on a cohort of 635 patients, revealing 83% sensitivity and 79% specificity, with an Area Under Curve (AUC) of 0.88. Importantly, analysis of the results revealed that 50% of the "false positive" cases originated from the newly HER2-low classification; therefore, the reported accuracy is even higher in real practice, considering the new HER2 low class (HER2 score 1&2-non-amp) that were recently approved by the FDA for targeted treatment.
These results present, to the best of our knowledge, the highest accuracy metrics of HER2 detection reported so far; furthermore, it emphasizes the fact that the AI model pointed to crucial biological features even before they were formally defined, such as the HER2-low classification. The ability to use image-based detection for HER2 inference may pose superior diagnostic and classification, especially in domains where human interpretation and subjective tests are being applied.
"AI-based image analysis algorithms show promise in identifying molecular characteristics of the tumor directly from the microscopic images. This requires simultaneous quantitative analysis of multiple morphological parameters and cannot be done by human pathologists," said Prof. Dov Hershkovitz, Director of the Pathology Dept., Sourasky Medical Center. "I believe Imagene's technology is going to change the way we practice pathology and perform biomarker analysis."
"Complex diseases with multiple genomic alterations such as cancer urges us to adopt advanced new analysis tools that can deal with such levels of complexity,'' said Dr. Nurit Paz-Yaacov, Chief Scientific Officer at Imagene. "We are grateful for the tight research collaboration with the Tel Aviv Sourasky Medical Center; by working together, we ensure that innovation and new technologies are meeting real clinical practice challenges."
Dr. Greenberg will present the research on September 6, 9:20 AM (CET). Imagene's scientific team and Sourasky Medical Center pathologists will also be available for individual presentations in booth #6, September 3-7 at the Congress Center, Basel, Switzerland.
Imagene's deep learning models have been shown to detect a wide range of cancer biomarkers in real-time, using only a digitally scanned Hematoxylin and Eosin (H&E) image. Imagene's molecular classifiers currently detect 28 different cancer biomarkers, spanning eight organs, including lung, thyroid, breast, ovary, CNS, bladder, colon, and hematologic malignancies. When compared to gold standard testing methods, Imagene's validation demonstrated, on average, sensitivity of 94.8%, greater than 90% specificity, and an AUC of over 0.95. Imagene has demonstrated the capability to identify a spectrum of alterations, including differential gene expression (e.g. HER2 presented here), mutations (e.g., EGFR), fusions and other structural variants (e.g., NTRK), and cancer signatures (e.g., HRD).
Imagene AI is a precision oncology diagnosis company. Its molecular and spatial intelligence platform delivers real-time biomarker reports using only digitized biopsy images, leading to faster diagnosis and better identification of treatment targets for patients. Its multidisciplinary team is composed of a diverse group of experts from the fields of science, medicine, and deep learning.
Imagene collaborates with top-tier medical centers and pharmaceutical companies worldwide, striving to render therapeutic decisions for cancer patients more accurate and accessible, profiling patients for clinical trials, and accelerating the drug development process.
For more information, visit Imagene-AI.com.
The mentioned products are currently in Research Use Only (RUO) and subject to regulatory clearances in the applicable markets.
Head of Marketing, Imagene