Researchers developed the crossNN AI model that classifies over 170 cancer types from DNA methylation data, achieving over 97% accuracy and enabling non-invasive diagnosis from liquid biopsies and tissue samples.
Key Details
- 1crossNN is a simple, explainable neural network AI trained on 8,000+ reference tumors and tested on 5,000+ tumors.
- 2Achieved 99.1% accuracy for brain tumor diagnosis; 97.8% accuracy across more than 170 tumor types from all organs.
- 3Uses DNA methylation profiles obtained from tissue or body fluids (e.g., cerebrospinal fluid), enabling some diagnoses to avoid surgical biopsies.
- 4Proven more accurate than previous AI solutions for tumor classification.
- 5The method is being prepared for clinical trials at all eight sites of the German Cancer Consortium.
- 6crossNN's workflow is fully explainable, meeting a key regulatory requirement for clinical adoption.
Why It Matters
crossNN demonstrates the potential of AI-driven, minimally invasive tumor diagnostics, which could complement imaging in cases where biopsy is too risky or impractical. Such models can improve diagnostic precision, reduce patient morbidity, and streamline selection of personalized treatments, impacting radiology workflows and multidisciplinary cancer care.
Source
EurekAlert
Related News
•EurekAlert
AI-Powered OCT Enables Rapid 'Optical Biopsy' for Early Endometrial Cancer Detection
A team at Washington University has developed a catheter-based 3D OCT system with AI to quickly and noninvasively detect early endometrial cancers.
•EurekAlert
AI Clinical Reasoning in Diagnostics and Digital Fatigue in Healthcare
Recent JMIR features explore large language models in clinical diagnostics and digital fatigue among healthcare professionals.
•EurekAlert
KAIST, MIT, Microsoft Develop Efficient AI Image Upsampling for Robotics
KAIST, MIT, and Microsoft have created 'Upsample Anything,' a training-free AI method to restore high-resolution visual data from compressed images with up to 16x improved GPU memory efficiency.