AI in Cancer Diagnosis and Molecular Testing Takes Center Stage at AMP 2026

Researchers are exploring how AI-based technologies can improve diagnostic accuracy, reduce unnecessary testing, and support more efficient clinical workflows.

Several studies being presented at the Association for Molecular Pathology (AMP) 2026 Annual Meeting & Expo highlight the potential of AI to improve molecular pathology while addressing challenges related to reliability, clinical validation, and expert oversight.

The AMP 2026 Annual Meeting & Expo, taking place November 10–14 at the Seattle Convention Center in Seattle, Washington, brings together molecular pathology professionals to discuss advances in diagnostic testing and precision medicine. Among the research being presented are studies examining AI-assisted cancer classification, automated genomic reporting, molecular testing optimization, and the prediction of genetic alterations from routine pathology images.

Researchers from ALMA Genomics, the University of Pennsylvania, the Massachusetts Institute of Technology, and collaborating institutions developed ALMA 3, an AI foundation model trained on DNA methylation data from more than 110,000 patient samples to classify hematolymphoid cancers.

In independent testing, the model detected blood and lymphatic tumors with 99.5% sensitivity and achieved a 90.9% positive predictive value when identifying specific cancer subtypes. Covering 81 molecular subtypes, including several that are difficult to distinguish using conventional methods, the approach could support more accurate cancer classification and help guide treatment decisions.

Researchers at the University of Michigan developed Turnkey Precision Oncology, an open-source, AI-assisted platform designed to analyze DNA and RNA sequencing data and generate evidence-supported molecular pathology reports. Its AI component, TPO-Agent, uses established cancer genetics databases and scientific literature to verify interpretations and reduce unsupported conclusions.

In an evaluation involving 50 genetic alterations across 10 patients, approximately 95% of the system's claims were supported by cited evidence, compared with 65% using a conventional AI approach. The broader platform also achieved approximately 95% agreement with clinical laboratory findings across 28 reports, suggesting that AI-assisted reporting could improve efficiency while maintaining expert oversight.

In another study, researchers at Memorial Sloan Kettering Cancer Center and the Icahn School of Medicine at Mount Sinai evaluated whether AI analysis of digital pathology slides could help prioritize rapid genetic testing in patients with lung adenocarcinoma. Six molecular pathology fellows reviewed 80 consecutive cases using AI-assisted assessments to determine whether immediate testing for EGFR mutations was necessary.

The approach identified 26 cases (33%) in which rapid EGFR testing could have been deferred without missing any confirmed EGFR-positive cases, potentially reducing redundant testing and preserving tumor tissue for comprehensive genomic profiling. However, the method was less reliable for determining when certain gene fusion tests could be deferred, highlighting the importance of further refinement and pathologist oversight.

Researchers at Beth Israel Deaconess Medical Center and Harvard Medical School investigated whether AI could distinguish hepatocellular carcinoma from intrahepatic cholangiocarcinoma by analyzing routinely collected pathology images. Using a foundation model capable of evaluating digital slides without additional disease-specific training, the team examined 50 tissue samples from The Cancer Genome Atlas.

The model identified visual features associated with cancer-specific gene expression, particularly markers characteristic of liver cells, with supporting evidence from RNA sequencing data. The findings suggest that AI-based image analysis could reveal molecular differences between liver tumors and provide additional information to support accurate cancer classification.

In a separate study, researchers at Stanford University evaluated whether AI could predict cancer-associated genetic mutations directly from digitized pathology slides of gastric tumors. The team analyzed tissue images from 367 patients using a pretrained AI foundation model to identify morphological patterns associated with alterations in four cancer-related genes.

The strongest predictive performance was observed for PIK3CA and CDH1 mutations, with area-under-the-curve scores of 0.80 and 0.77, respectively. Although further validation is needed, the findings suggest that AI-assisted pathology could eventually help identify patients most likely to benefit from additional molecular testing, particularly in settings where genomic sequencing is not readily available.

Together, the studies presented at AMP 2026 highlight the expanding role of AI in molecular diagnostics, from improving the classification of hematologic malignancies and interpreting cancer genomic data to optimizing testing strategies and extracting molecular information from conventional pathology images.

The findings demonstrate how AI technologies could complement established diagnostic methods, improve laboratory efficiency, and support more personalized cancer care, while underscoring the need for continued clinical validation and human oversight.

Related Links
Association for Molecular Pathology