Modern Pathology最新文献

Accurate histopathological evaluation of pancreatic ductal adenocarcinoma (PDAC), including primary tumor lesions and lymph node metastases, is critical for prognostic evaluation and personalized therapeutic strategies. Distinct from other solid tumors, PDAC presents unique diagnostic challenges due to its extensive desmoplasia, unclear tumor boundary, and difficulty in differentiation from chronic pancreatitis. These characteristics not only complicate pathological diagnosis but also hinder the acquisition of pixel-level annotations required for training computational pathology models. Here, we present PANseg, a multi-scale weakly supervised deep learning framework for PDAC segmentation, trained and tested on 368 whole-slide images (WSIs) from 192 patients across two independent centers. Utilizing only image-level labels (2,048×2,048 pixels), PANseg achieved comparable performance to fully supervised baseline (FSB) across the internal test set 1 (17 patients/58 WSIs; PANseg AUROC: 0.969 vs FSB AUROC: 0.968), internal test set 2 (40 patients/44 WSIs; PANseg AUROC: 0.991 vs FSB AUROC: 0.980) and external test set (20 patients/20 WSIs; PANseg AUROC: 0.950 vs FSB AUROC: 0.958). Moreover, the model demonstrated considerable generalizability with previously unseen sample types, attaining AUROCs of 0.878 on fresh-frozen specimens (20 patients/20 WSIs) and 0.821 on biopsy sections (20 patients/20 WSIs). In lymph node metastasis detection, PANseg augmented diagnostic accuracy of six pathologists from 0.888 to 0.961, while reducing average diagnostic time by 32.6% (72.0 vs 48.5 minutes). This study demonstrates that our weakly supervised model can achieve expert-level segmentation performance and substantially reducing annotation burden. The clinical implementation of PANseg holds great potential in enhancing diagnostic precision and workflow efficiency in routine histopathological assessment of PDAC.

Mucinous cystic neoplasms (MCNs) are precursors to invasive pancreatic cancer. Conflicting clinical management recommendations call for better molecular characterization to improve our understanding of their tumorigenesis and risk assessment. We sampled epithelial tissue from a total of 18 surgically resected MCNs and performed dedicated multi-region analysis of somatic genetic alterations by targeted next-generation sequencing of 25 driver genes. In addition, we performed whole exome sequencing and immunohistochemistry on selected samples to supplement our analyses. In total, 128 samples of epithelial MCN tissue were sequenced and analyzed, including samples from 13 small MCNs with low-grade dysplasia, one small MCN with high-grade dysplasia, and four large MCNs with high-grade dysplasia. Eight of 13 (61.5%) comprehensively sampled small low-grade MCNs lacked somatic driver gene mutations in all tissue blocks. These MCNs were lined by predominantly flat epithelium. In contrast, the majority of MCNs with driver gene mutations were predominantly lined by mucin-rich epithelium. No heterogeneity in KRAS mutations was seen across the sampled regions. Multi-region genetic analysis of four large MCNs with high-grade dysplasia provide insights into neoplastic progression, with shared somatic alterations suggesting that high-grade dysplasia arises from low-grade mucin-rich epithelium. These findings were supported by complementary whole exome sequencing studies in 26 MCN epithelium samples. The neoplastic epithelium in the majority of small MCNs does not harbor somatic mutations in pancreatic driver genes. The genetic findings from multi-region analysis on MCNs contrast previous investigations in other mucin-producing pancreatic cysts, indicating distinct mechanisms in early tumorigenesis. This calls for a more nuanced risk assessment in MCNs, requiring improved pre-operative assessment tools.