Introduction/Background Tertiary lymphoid structures (TLSs) are known to be a marker of peripheral inflammation in several cancer types, however there is no evidence of clinical benefit of immune checkpoint inhibitor (ICI) and special interplay pattern of TLSs in endometrial cancer.

Methodology We developed an artificial intelligence (AI)-based TLSs detection program using transfer learning DeepLabV3 and performed spatial analyses of TLSs in 958 tiles from tumor samples of 258 endometrial cancer patients. And we applied this AI-based program to evaluate the relationship between spatial distribution (according to distance from tumor burden) of TLSs and survival rate or antitumor effect of immune checkpoint inhibitors for endometrial cancer patients.

Results First, we could make a program that automatically recognized TLSs in tumor samples with an accuracy agreement rate of 96% for training data and 92% for evaluation data. In 104 patients with endometrial cancer, TLSs were detected in 81 patients (78%), and the patients with TLSs at >500um from tumor burden (extra-TLSs) were closely related to favorable progression free survival (P <0.004), while the patients with ≤500um from tumor burden (peri-TLSs). Besides, among 12 endometrial cancer patients treated with ICI, 4 of 5 patients with extra-TLSs showed clinical response (80% of response rate [RR]: 2 complete response and 2 partial response [PR] ), while only 1 of 7 patients with peri-TLSs showed clinical response (RR 14%: 1 PR).

Conclusion AI-based spatial anaysis of TLSs may be useful to predict the prognosis and one of anti-tumor biomarker of ICI in advanced endometrial cancer.

Disclosures Spatial distribution of TLSs may be closely related to patients‘ survival, and extra-TLSs may represent local immune status in tumor microenvironment.