Studies demonstrate artificial intelligence achieving 93% cancer diagnostic accuracy and enabling noninvasive tumor analysis through spinal fluid testing.
The Association for Molecular Pathology (AMP) 2025 Annual Meeting & Expo is featuring multiple studies highlighting artificial intelligence (AI) applications in molecular diagnostics, with research demonstrating enhanced accuracy and speed in cancer diagnosis and chromosomal analysis.
The meeting, taking place Nov 11-15 in Boston, is showcasing AI developments that address current diagnostic challenges, from improving RNA sequencing accuracy to enabling noninvasive tumor analysis.
AI Classifier Achieves High Cancer Diagnostic Accuracy
Researchers from The Hospital for Sick Children developed a web platform incorporating an AI classifier designed to integrate RNA sequencing into clinical workflows while handling heterogeneous datasets. The model achieved 93% diagnostic accuracy on cancer subtypes covered by the platform.
The system adapts and incorporates new subtypes, increasing accuracy with each new sample. Researchers aim to cover new subtypes with only five reference samples and scale up the platform to include more diverse benign and malignant entities.
The goal for the platform is to cover new subtypes with only five reference samples, according to Pedro Lemos Ballester, PhD, at The Hospital for Sick Children and the study’s lead author, who will present the work during a poster session at 9:15 am on Saturday, Nov 15.
Noninvasive CNS Tumor Diagnosis Using AI
Researchers at Soonchunhyang University in South Korea created two AI models to classify central nervous system tumor samples using cerebrospinal fluid-derived circulating tumor DNA as a noninvasive alternative to tissue biopsies.
The team developed a dense neural network trained on mutation data from 12 key genes via next-generation sequencing and a convolutional neural network trained on standardized MRI images. Combining outputs from both models improved prediction and classification accuracy.
This approach enables accurate mutation prediction and treatment planning before surgery, allowing surgeons to anticipate tumor biology rather than wait for postoperative tissue analysis.
Jieun Kim, MD, PhD, from Soonchunhyang University will present this work during a poster session at 9:15 am on Saturday, Nov 15.
AI Reveals Chromosomal Changes in Blood Cancer
Wake Forest University School of Medicine researchers used an AI-trained karyotyping algorithm to analyze chromosomal abnormalities in GATA2 deficiency syndrome-related leukemia. GATA2 deficiency syndrome is a rare autosomal dominant genetic disorder causing predisposition to immunodeficiency and myeloid malignancy.
The AI system enabled rapid generation and review of hundreds of images, improving detection and confidence in identifying complex clonal chromosome rearrangements. The AI-assisted karyotyping revealed detailed clonal evolution over time in a patient’s acute myeloid leukemia, capturing multiple chromosomal changes that tracked disease progression.
Lynne Rosenblum, PhD, at Wake Forest University School of Medicine will present this research during a poster session at 9:15 am on Saturday, Nov 15.
Framework for Personalized Oncology Care
Augusta University researchers created a computational framework to train and compare AI models that predict genomic and transcriptomic information directly from hematoxylin and eosin-stained slide images. The framework successfully compared different AI models and handled prediction tasks including identifying gene activity and prognosis.
When patient clinical information was incorporated, predictions became more informative. The researchers found that different AI models perform differently depending on the specific goal or data type, requiring future standardization of testing and benchmarks.
Pankaj Ahluwalia, PhD, presented this work overseen by Ravindra Kolhe, MD, PhD, at Augusta University during a poster session at 9:15 am on Friday, Nov 14.
ID 348741311 | Artificial Intelligence © Wave Break Media Ltd | Dreamstime.com
