By Lynda Williams, medwireNews Reporter
medwireNews: Artificial intelligence (AI) can expedite assessment of tumour-infiltrating lymphocytes (TILs) and immune phenotype (IP) identification in resected pancreatic ductal adenocarcinoma (PDAC), research demonstrates.
“Results of this cohort study suggest that the use of AI has markedly condensed the labor-intensive process of TIL assessment, potentially rendering the process more feasible and practical in clinical application”, say Joo Kyung Park (Samsung Medical Center, Sungkyunkwan University School of Medicine, Seoul, South Korea) and co-workers.
“Importantly, the IP may be one of the most important prognostic biomarkers in resected PDACs”, they write in JAMA Surgery.
Between 2017 and 2020, hematoxylin and eosin-stained slides of PDAC were collated from 304 patients (56.3% men, average age 66.8 years) with clinical stage I (54.3%) or stage II (45.7%) disease.
All patients had undergone an R0 resection and were followed up for a median 35.0 months, during which time 188 patients experienced recurrence and 190 patients died (27 without recurrence). The median recurrence-free survival (RFS) duration was 15.80 months and the median overall survival (OS) was 35.87 months.
The researchers trained a deep learning-based AI model to distinguish between tumour and stromal cell compartments on the slides and quantify lymphocytes.
Using the AI system, 9.9% of the tumours were classified as having an immune-inflamed phenotype (IIP), with high levels of lymphoid cells in the tumour area. The remaining tumours were classified as having either an immune-excluded phenotype (IEP, 85.2%), with low levels of lymphoid cells in the tumour area but high levels in the stroma, or an immune-desert phenotype (IDP, 4.9%), with low levels of lymphoid cells in both areas.
Patients with IIP had significantly better RFS than those with IEP or IDP (median unreached vs 14.63 and 6.57 months, respectively) and this was also true for OS (median unreached vs 35.11 and 11.6 months, respectively).
In addition, patients whose intratumoural TIL density was in the highest quartile had significantly better RFS and OS than those whose density was in a lower quartile, at a median 21.67 versus 13.55 months and 52.47 versus 32.83 months, respectively. By contrast, no significant difference in survival was found by stromal TIL density, the researchers say.
The team also found that RFS and OS differed significantly between the IP groups when patients who had and had not received adjuvant treatment were assessed separately, with the IIP group again having “the most favorable prognosis” regardless of adjuvant therapy receipt.
IEP patients who received adjuvant therapy had better survival than those who did not, with median OS of 44.05 versus 15.9 months, which the researchers say indicates that this subgroup “may benefit from adjuvant therapy.”
As expected, pathological stage also significantly predicted prognosis, with median OS ranging from 54.43 months among those with stage I disease to 32.61 months for stage II patients and 17.33 months for stage III patients.
However, when pathological stage was combined with IP, Park et al report that “[r]emarkably, patients with stage II and IIP tended to have longer OS than those with stage I and non-IIP”, with median OS of unreached and 52.47 months, respectively.
“Future research should aim to evaluate the utility of this technology in patients receiving anticancer therapy, including those with advanced PDAC, and extend its application to biopsy specimens, which could enable preoperative classification of the immunophenotype and broader clinical integration”, the authors therefore suggest.
In addition to needing to validate these findings in a large study population, the team concludes that “immunophenotype classification strategies, including cutoff thresholds, the use of categorical vs continuous scoring, and the potential integration with molecular subtypes, may require refinement as more PDAC-specific data become available.”
Discussing the findings in a linked comment, Syed Ahmed and colleagues, at the University of Cincinnati in Ohio, USA, believe the study shows that the technique “can provide robust prognostic stratification” but observe that “the current analysis does not distinguish lymphocyte subtypes, such as myeloid-derived suppressor cells or regulatory T cells, or quantify spatial proximity between immune components and tumor cells.”
They suggest: “This type of information could easily be ascertained by similar AI harnessed processes and could help inform treatment decisions.”
medwireNews is an independent medical news service provided by Springer Healthcare Ltd. © 2025 Springer Healthcare Ltd, part of Springer Nature
JAMA Surg 2025; doi:10.1001/jamasurg.2025.1999
JAMA Surg 2025; doi:10.1001/jamasurg.2025.1978
https://pubmed.ncbi.nlm.nih.gov/40560550/
https://pubmed.ncbi.nlm.nih.gov/40560593/
AI analysis aids pancreatic cancer prognostication
