Detecting Extrachromosomal DNA from Routine Histopathology.

Area of Science:

• Oncology
• Genomics
• Computational Pathology

Background:

• Extrachromosomal DNA (ecDNA) drives oncogene amplification, tumor heterogeneity, and poor outcomes.
• Current ecDNA detection methods require specialized genomic assays, limiting routine diagnostic use.

Purpose of the Study:

• To develop a method for inferring ecDNA status directly from standard histopathology images.
• To enable scalable screening of tumors for ecDNA amplification using routine diagnostics.

Main Methods:

• Development of an end-to-end, weakly supervised deep learning framework.
• Aggregation of thousands of high-magnification image patches per slide with augmentation and attention.
• Validation across twelve cancer types from The Cancer Genome Atlas.

Main Results:

• The framework successfully identifies tumors with genomic amplifications from standard pathology slides.
• It distinguishes ecDNA-amplified tumors from chromosomally amplified or non-amplified ones, particularly in glioblastoma.
• Attention maps highlight regions with altered nuclear chromatin intensity and texture, correlating with ecDNA status and survival.

Conclusions:

• Histomorphologic features indicative of ecDNA amplification are detectable in routine pathology images.
• This approach allows for scalable screening to prioritize tumors for molecular testing.
• Deep learning on histopathology images offers a promising avenue for ecDNA detection in clinical settings.