Editors: Rouzan Karabakhtsian, MD, PhD, professor of pathology and director of the Women’s Health Pathology Fellowship, Albert Einstein College of Medicine, Montefiore Medical Center, Bronx, NY; S. Emily Bachert, MD, associate pathologist, Brigham and Women’s Hospital, Boston; Amarpreet Bhalla, MD, assistant professor of pathology, Albert Einstein College of Medicine, Montefiore Medical Center; Divya Sharma, MD, associate professor, Department of Pathology and Laboratory Medicine, University of Cincinnati Medical Center; and Paula Toro, MD, gastrointestinal and hepatobiliary fellow, Cleveland Clinic.
Use of AI to define spatial patterns of tumor-infiltrating lymphocytes associated with patient outcome
May 2026—Gastrointestinal cancers, including those of the esophagus, stomach, colon, rectum, pancreas, and liver, account for more than 25 percent of all cancer diagnoses and 35 percent of cancer-related fatalities worldwide. Quantification of tumor-infiltrating lymphocytes (TILs) is a prognostic marker of cancer. The authors conducted a study in which they used computer vision and machine-learning approaches to evaluate the prognostic significance of computational pathology features relating to spatial arrangement and diversity in the appearance of TILs and cancer nuclei across five types of GI cancers—colon, stomach, and pancreatic cancer; rectal adenocarcinoma; and liver hepatocellular carcinoma. The study comprised more than 1,700 patients from four sites. Pathomic features (2,236) were extracted from hematoxylin and eosin-stained whole slide images, and the top nine features were selected by the Least Absolute Shrinkage and Selection Operator (LASSO) Cox model. The top prognostic features identified were related to the spatial relationships between TILs and the closest cancer nuclei and to features of tumor nuclei shape and texture captured within local cellular clusters. The trained model used in the study identified that low-risk patients have significantly better overall survival than high-risk patients, with a hazard ratio (HR) of 2.28 (95 percent confidence interval [CI], 1.32–3.93; P = .0032) in liver hepatocellular carcinoma, HR of 2.79 (95 percent CI, 1.66–4.68; P = .0001) in pancreatic adenocarcinoma, HR of 5.85 (95 percent CI, 2.53–15.5; P = .0002) in rectal adenocarcinoma, and HR of 1.81 (95 percent CI, 1.07–3.07; P = .0268) in gastric adenocarcinoma. With regard to external validation sets of colorectal cancer patients, the study model yielded an HR of 2.32 (95 percent CI, 1.67–3.23; P < .0001) in The Cancer Genome Atlas–Colon Adenocarcinoma (TCGA-COAD) data set, HR of 2.32 (95 percent CI, 1.67–3.23; P < .0001) in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO)–COAD, and HR of 3.38 (95 percent CI, 1.99–5.71; P < .0001) in the Emory data set. Multivariable survival analysis showed that the trained model was prognostic independent of stage, age, race, and sex. The authors’ findings suggest that the spatial relationships of TILs and cancer nuclei are prognostic of survival across multiple GI cancer types.
Chen C, Mejbel HA, Pathak T, et al. Artificial intelligence defines spatial patterns of tumor-infiltrating lymphocytes highly associated with outcome—a pan-GI cancer study. ESMO Open. 2025. doi.org/10.1016/j.esmoop.2025.105757
Correspondence: Dr. Anant Madabhushi at anantm@emory.edu
Expanding the clinicopathologic spectrum of EWSR1::SSX-rearranged sarcomas
Gene rearrangements involving EWSR1 or SSX genes are known to play a role in sarcomagenesis. Yet, sarcomas harboring EWSR1::SSX fusions are rare. To better understand tumors associated with this distinctive genetic event, the authors studied 11 EWSR1::SSX sarcomas, affecting 10 females and one male (average age, 46 years; range, 22–72 years). Eight tumors arose in bone (rib, femur, pelvis, and vertebra, with multifocal bone involvement at presentation in three cases) and two in soft tissue (deep thigh and groin). One patient presented with a visceral (lung) mass. The tumors were bulky and averaged 12.1 cm (range, 4–20 cm). Histologically, in keeping with a translocation-driven sarcoma, all tumors were cytologically monotonous. Seven tumors were osteosarcomas, six of which were classified as sclerosing osteosarcomas with extensive production of bone matrix. Three tumors were composed of uniform fascicles of spindle cells, punctuated in two cases by a biphasic glandular or nested epithelioid component, reminiscent of synovial sarcoma. One tumor was an undifferentiated sarcoma with round to spindle cell cytomorphology and focal osteogenic differentiation. Using immunohistochemistry, five of five cases tested were positive for SSX C-terminus and one of five showed patchy weak staining with SS18::SSX fusion-specific antibody. All cases tested for CD99 (four), TLE1 (three), and BCOR (one) were positive. All cases underwent next-generation sequencing, and all tumors harbored EWSR1::SSX fusions, including EWSR1::SSX1 (n = 7), EWSR1::SSX2 (n = 2), EWSR1::SSX3 (n = 1), and a novel EWSR1::SSX4 fusion (n=1). Follow-up was available for nine patients. Five patients died of disease 1.5 to 14 years (average, six years) after diagnosis, while two patients were alive with metastatic disease, one was alive without disease at 25 months, and one patient presented recently. The authors concluded that sarcomas with EWSR1::SSX fusions are rare and clinically aggressive. The histologic patterns in this series and in prior reports are heterogeneous and essentially consist of primitive round or epithelioid cells, osteoblasts that produce bone, or uniform small spindle cells arranged in tight fascicles, as seen in synovial sarcoma.
Gross JM, Suster DI, Zou Y, et al. Expanding the clinicopathologic spectrum of EWSR1::SSX-rearranged sarcomas: Series of 11 cases including osteosarcomas and a novel EWSR1::SSX4 fusion. Mod Pathol. 2026. doi.org/10.1016/j.modpat.2025.100922
Correspondence: Dr. Gregory W. Charville at gwc@stanford.edu