30-Second Takeaway
- AI and liquid biopsy tools are correcting misclassification and accelerating diagnosis across solid and hematologic malignancies.
- Off-target NGS reads and methylation profiling now deliver near–array-level CNS tumor copy-number and epigenetic resolution.
- Routine H&E slides increasingly encode actionable molecular subtype information, particularly in prostate cancer.
- Targeted IHC, especially MLH1 in SSLs, improves detection of clinically significant precursors without broad panel overuse.
- Highly sensitive ctDNA and colorimetric miRNA platforms foreshadow low-cost, scalable disease monitoring and point-of-care diagnostics.
Week ending March 21, 2026
Emerging diagnostic tools reshaping pathologic classification and molecular risk stratification
AI tissue-of-origin model corrects a meaningful fraction of presumed lung squamous cell carcinomas
An AI tissue-of-origin model (GPSai) reclassified 3.1% of 3,958 tumors submitted as presumed primary lung squamous cell carcinoma. Most reclassified cases were actually cutaneous, orogenital, urothelial, or thymic squamous carcinomas, or other rare primaries. Orthogonal confirmation used targeted IHC panels, HPV status, UV mutational signatures, and clinical history or imaging findings. In 71.5% of reclassified cases, the new diagnosis changed guideline-preferred first-line systemic therapy options. These data support integrating AI-based tissue-of-origin review into molecular workflows for difficult thoracic squamous cases.
Off-target reads from small CNS NGS panels provide reliable genome-wide CNV profiles
Genome-wide CNV profiles derived from off-target reads of a <0.2 Mb CNS hybrid-capture panel showed 95% concordance with methylation array arm-level calls. All focal amplifications, including clinically relevant EGFR, MDM4, and MYCN events, and most homozygous CDKN2A/B deletions were captured. In meningiomas, NGS-derived CNVs supported WHO-relevant molecular upgrading in 16% of histologically lower grade tumors. Focal 7q CNVs compatible with BRAF fusions were observed in most pilocytic astrocytomas known to carry BRAF fusions. These findings suggest many neuropathology labs can obtain actionable CNV data from existing small NGS panels without additional assays.
AI predicts prostate cancer PAM50 and PSC subtypes directly from routine biopsy H&E slides
An AI framework (UNIv2-MIL) predicted PAM50 and PSC molecular subtypes from standard prostate biopsy H&E slides with AUCs of 0.863 and 0.81. High-attention regions revealed reproducible histologic patterns associated with specific molecular subtypes, supporting biological plausibility. In an independent cohort, AI-predicted luminal subtypes showed better response to hormone therapy, with statistically significant associations. In another cohort transitioning from active surveillance to prostatectomy, AI-predicted luminal B and luminal proliferating scores correlated with adverse pathologic features. These data suggest slide-only AI models may eventually guide risk stratification and systemic therapy decisions without upfront molecular testing.
Liquid biopsy rapidly and accurately diagnoses EBV-positive Burkitt lymphoma in endemic settings
A liquid biopsy model combining clinical features, MYC alterations, and EBV fragmentomics diagnosed EBV-positive Burkitt lymphoma with AUC 0.95 in a multicenter East African cohort. Sensitivity and specificity were 0.86 and 0.95, respectively, and performance was confirmed in an external validation set with AUC 0.98. Head-to-head comparison in 58 participants showed liquid biopsy cut median diagnostic turnaround from 46.8 days to 6.5 days. Liquid biopsy was the only diagnostic result available at multidisciplinary review in 42% of participants, effectively driving treatment decisions. These results support integrating ctDNA-based assays into diagnostic pathways where tissue access and hematopathology capacity are limited.
References
Numbered in order of appearance. Click any reference to view details.
Additional Reads
Optional additional studies from this edition.