Ibex Medical Analytics (Ibex), the leader in AI-enabled cancer diagnostics, and the Institut Curie, France’s leading cancer center, today announced excellent results in a clinical study validating and evaluating the clinical use of Ibex’s Galen™ Breast. The study, conducted at the Institut Curie in France and Maccabi Healthcare Services in Israel, was published in Nature’s journal npj Breast Cancer1.
Prof. Anne Vincent-Salomon from the Institut Curie and Dr. Study led by Judith Sandbank of Maccabi Healthcare Services2 is the first to report an AI-based algorithm that can accurately detect such a wide range of clinically significant pathological features in breast biopsies. In addition, the study reports on the first-ever implementation of such an AI solution in routine clinical use in a pathology laboratory and demonstrate its clinical utility as a decision-support tool that helps pathologists reduce diagnostic errors and improve diagnostic quality.
“This publication is a significant addition to the impressive body of evidence we have produced while studying Ibex’s AI solution over the past few years, showing consistently high performance,” said Anne Vincent-Salomon, MD, Head of the Pathology Department at Institut Curie and Professor at the University Paris-Sciences et Lettres. “We are delighted with our fruitful collaboration with Ibex, which has given Institut Curie pathologists first-hand experience with this AI tool, a platform that has undeniably proven value in their daily clinical practice, enabling us to optimize diagnoses, accelerate treatment decisions and ultimately improve care for our patients.”
With more than 2.2 million new cases per year, breast cancer is the most common malignancy worldwide3. Therefore, accurate and timely diagnosis is key to making treatment decisions by oncologists and improving patient survival rates. In recent years, rapid advances in personalized medicine have led to the increasing complexity of cancer diagnosis. These trends, combined with an increase in the overall incidence of breast cancer and a worldwide decline in the number of pathologists, have resulted in an increased workload in pathology departments. There is clearly a growing need for automated decision support solutions and tools that help pathologists detect cancer faster and with the highest accuracy.
Ibex developed Galen Breast to help pathologists diagnose breast biopsies by providing insights that help identify and categorize different types of invasive and non-invasive breast cancer, as well as other clinically significant pathological features. The solution’s AI algorithm was trained to identify more than 50 breast-specific features that may appear in breast biopsies using advanced deep-learning technologies applied to hundreds of thousands of image samples.
“I was impressed with the study outcomes, the very high accuracy levels and the breadth of detection capabilities offered by Ibex’s AI technology, all on par with the performance of expert pathologists,” said Stuart Schnitt, MD, Chief of Breast Oncologic Pathology at the Dana-Farber/Brigham and Women’s Cancer Center and Professor of Pathology at Harvard Medical School and co-author on the study4. “It is exciting to take part in investigating and validating new innovations which are going to re-shape our practice for years to come and I look forward to see more AI applications roll out into routine clinical use as they demonstrate their clinical validity.”
The study included 841 blinded full-frame images from 436 breast biopsies stained with either H&E or HES and digitized with different scanning systems. The images were analyzed using Ibex’s Galen breast solution and the output of the AI algorithm was assessed using a blinded consensus diagnosis by two breast pathologists. The AI algorithm was able to identify invasive carcinomas, including several rare specific types of breast cancer, with exceptionally high accuracy, regardless of staining protocol and scanner type. Specifically, the algorithm accurately identified invasive lobular carcinomas from invasive carcinomas with no specific type, as well as rare types (metaplastic or mucinous carcinomas) and ductal carcinoma in situ and atypical ductal hyperplasia (DCIS and ADH), and different classifications of carcinoma in situ (DCIS high/intermediate-grade vs. low-grade/ADH). The algorithm also demonstrated high accuracy in identifying key prognostic factors such as tumor infiltrating lymphocytes (TILs) and angiolymphatic invasion, as well as non-cancerous features such as columnar cell changes and microcalcifications.
“We are proud of the study outcomes, which demonstrate the robustness of our breast algorithm across an unprecedent gamut of cancer types and other clinically important pathologies,” said Dr. Manuela Vecsler, Director of Clinical & Scientific Affairs at Ibex Medical Analytics. “The detection of more than 50 breast features renders the AI algorithm more comprehensive, accurate, and explainable, and thus Galen Breast has the ability to support pathologists across a wider range of tasks, as attested by the increasing number of laboratories that deploy the solution in routine use.”
