Expanding cancer diagnostics through an analysis of lymph nodes

Clinical Deployment and Validation of a Multi-Cancer Deep Learning Algorithm for Lymph Node Metastasis Detection Based on Foundation Models

This study is based on a tight collaboration between computer science and clinical application. The detection of cancer cells in a patient’s lymph nodes is key to the clinical management of cancer. Analyzing the status of lymph nodes requires a pathologist’s skill, yet it is labor-intensive and time-consuming. To support pathologists in this task, a computer-assisted diagnostic tool that uses deep learning for the detection of colorectal lymph node metastases was developed. Now the scientists will extend the training of this algorithm to lymph nodes from 10 cancer types to expand its utility for daily use in hospital oncology.

What bone marrow imaging can tell us about leukemia

Deepmarrow: Investigating Marrow Remodeling upon Intensive Chemotherapy as a Predictor of Response in Acute Myeloid Leukemia

In acute myeloid leukemia (AML), changes in the zone surrounding the tumor, called the stroma, can convert zones of rapid blood cell production into sites of pathological cell growth, and further lead to overt leukemia.  The goal of this study is to develop digital machine learning tools based on blood cell pathology, in order to quantify components of the connective tissue in bone marrow. This tool may also help identify AML patients with a high risk of relapsing. Such a tool should help predict and prevent relapse in AML—a clearly unmet clinical need.

Diagnostics of a rare lymphoma through integrative AI

Transforming the Diagnostics of Nodal Marginal Zone Lymphomas by Integrating Digital Pathology, Molecular Biology, and Artificial Intelligence

Nodal marginal zone lymphoma (NMZL) is a rare and difficult to diagnose B-cell malignancy that is often wrongly diagnosed. The applicants propose a digital pathology approach that they hope surpasses expert pathologists in correctly diagnosing NMZL. They will use deep learning to integrate imaging, clinical and molecular data, generating a classifier that will allow clinicians to upload their whole slide scans and get a correct NMZL diagnosis. Based on 900 cases, samples and data obtained across Europe, they hope to provide a clinically applicable, yet technically sophisticated way to identify a rare lymphoma subtype.

Spatial Biology for Risk Stratification of Colorectal Cancer

Stratifying colon cancer risk through imaging and transcription profiles

In order to identify prognostic biomarkers for stage II colorectal cancers (CRC), the investigators will use computational approaches to combine advanced molecular ‘omics data with histological images from 1800 past CRC patients. The goal will be to identify prognostic markers in networks, and to validate these on new cohorts of patients. This is an advanced digital pathology project.

Does tissue fibrosis trigger brain cancer recurrence?

Spatial Analysis and Functional Interrogation of Fibrotic Scarring Niches in Glioblastoma Recurrence

Zones of fibrotic scarring are associated with tumor relapse in aggressive brain tumors. This study will use cutting-edge spatial multi-omics technologies and digital pathology analyses to explore the mechanisms that generate fibrosis in glioblastoma patients. Ultimately, the scientists hope to uncover novel therapeutic targets that disrupt the tumor-protective fibrotic niche and thereby prevent disease recurrence.

Distinguishing primary tumors from metastases in lung with AI

Multiple Lung Tumor Nodules: Establishing Cost-Effective Molecular and Digital Pathology Methods to Separate Multiple Primary Lung Cancers from Intrapulmonary Metastases

Analyzing stained tumor biopsies by microscopy is the pathologist’s craft. Now, this can be enhanced and augmented by computer-based analysis and integration with other information.

Using a large data base of tissue sections from primary lung adenocarcinoma, the team will compare these with metastatic nodules in the lung derived from other primary tumors in order to optimize a new digital pathology tool (DPLAS) that will stratify lung cancers and allow patient triaging. A strong feature of the project is that the technology is based on routinely available H&E slides, which enables broad usage of the tool.