Kang, Jeonghyun

Abstract Histological distribution and density of tumor infiltrating lymphocytes (TILs) are known to be informative to cancer patients’ outcome. However, due to the heterogeneity and huge size of digitized pathology slides, it has been a challenging problem to quantify TILs-related features in whole slide images (WSI). This study aimed to explore the impact of spatial distribution of TILs quantified by deep learning (DL) approaches based on digitalized WSI stained with hematoxylin and eosin in patients with colorectal cancer (CRC). The cancer and TILs regions are first separately identified by deep learning models in the WSI. The TILs densities at tumor invasive margins are then quantified by image analysis algorithms and used for prognosis of progression-free survival (PFS) in CRC patients. The prognostic impact of spatial distributions of TILs in CRC patients was explored in the Yonsei cohort (n=180) and validated in the TCGA cohort (n=268). Two experienced pathologists manually measured TILs at the most invasive margin as 0-3 by the Klintrup-Mäkinen (KM) grading method and compared to DL approaches. On multivariate analysis, TILs densities within 200µm of the invasive margin (f_im200) was remained as the most significant prognostic factor for PFS (HR 0.004 [95% CI, 0.0001-0.1502], p=.002) in the Yonsei cohort. On multivariate analysis using the TCGA dataset, f_im200 retained prognostic significance for PFS (HR 0.031, [95% CI 0.001-0.645], p=.024). Inter-observer agreement of manual KM grading was insignificant in both cohorts. The survival analysis by pathologists’ manual KM grading showed statistically significant different PFS from the TCGA cohort, but not the Yonsei cohort. In summary, automated quantification of TILs at the invasive margin showed a prognostic utility to predict PFS, and could provide reproducible TILs density measurement in CRC patients. Citation Format: Hongming Xu, Yoon Jin Cha, Jean Clemenceau, Jinhwan Choi, Sung Hak Lee, Jeonghyun Kang, Tae Hyun Hwang. Spatial analysis of tumor infiltrating lymphocytes in histological images using deep learning predicts progression-free survival in colorectal cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 642.

Abstract Background: Microsatellite Instability (MSI) is a prognostic and predictive biomarker which can guide treatments including immunotherapy for patients with gastrointestinal cancer. Universal MSI testing would benefit to patients to receive a better therapy, but many patients remain untested. To deliver broadly accessible MSI testing, deep learning models have demonstrated their feasibility for MSI prediction using H&E-stained whole-slide images (WSIs). However, to be deployed into clinical routine care, prediction models should be validated in datasets from multicenter and multiethnic groups. In addition, these models should provide uncertainty of prediction to help clinicians to make informed decisions. Method: We develop a MSI prediction model using WSIs based on Deep Gaussian process (DGP), a Bayesian model being able to model the uncertainty in prediction. We implement a DGP model in the transfer learning, where a WPI is decomposed into multiple non-overlap image patches which are converted to feature vectors by a pretrained convolutional neural network (CNN). Then, the DGP model makes a prediction for an image by averaging score function values at all the feature vectors in the image in weakly supervised learning (WSL). To test our method, we collected H&E stained colorectal/stomach cancer slides (n=1,619) from multi-institutions and multiethnic groups, including TCGA Colorectal (CRC; n=351) and Stomach Adenocarcinoma (STAD; n=174) and six datasets from tertiary hospitals in Korea (i.e., Yonsei-1 (n=174), Yonsei-1-remade (n=146), Yonsei-2 (n=95), St. Mary-1 (n=48), St. Mary-2 (n=50), and Yonsei-CLASSIC trial (n=581)). We also compare our method to multiple deep learning methods, including Densenet, Resnet, Shufflenet, Googlenet, Squeezenet and two WSL models, MIL (Campanell et al., 2019) and CLAM (Lu et al., 2021). Result: In each cancer type, we train each model using TCGA dataset as discovery cohort and tested on the rest of datasets as validation cohort. We use the area under of roc curve (AUC) for the evaluation metric. For the CRC cohorts, our DGP method achieves overall best performance with 0.81 (Yonsei-1), 0.82 (Yonsei-1-remade), 0.89 (Yonsei-2), 0.85 (St. Mary-1) 0.75 (St. Mary-2) AUCs compared to the state-of-the art methods. For the STAD cohorts, our method also achieves overall best performance with 0.74 AUC (Yonsei-classic). In addition, incorporating the uncertainty in prediction measured by our method improves the model’s performance to predict MSI. Specifically, removing cases with prediction results with high uncertainty, which could lead false positive prediction, significantly improves MSI prediction performance. Taken together, these results demonstrate the robustness and generalisability of the DGP model for MSI prediction across multicenter and multiethnic datasets. Citation Format: Sunho Park, Hongming Xu, Sung Hak Lee, Jeonghyun Kang, Tae Hyun Hwang. Deep Gaussian process with uncertainty estimation Improves microsatellite instability prediction based on whole slide image: A retrospective multicenter and multiethnic cohort study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5012.

Abstract This study aimed to explore the prognostic impact of spatial distribution of tumor‐infiltrating lymphocytes (TILs) quantified by deep learning (DL) approaches based on digitalized whole‐slide images stained with hematoxylin and eosin in patients with colorectal cancer (CRC). The prognostic impact of spatial distributions of TILs in patients with CRC was explored in the Yonsei cohort ( n = 180) and validated in The Cancer Genome Atlas (TCGA) cohort ( n = 268). Two experienced pathologists manually measured TILs at the most invasive margin (IM) as 0–3 by the Klintrup–Mäkinen (KM) grading method and this was compared to DL approaches. Inter‐rater agreement for TILs was measured using Cohen's kappa coefficient. On multivariate analysis of spatial TIL features derived by DL approaches and clinicopathological variables including tumor stage, microsatellite instability, and KRAS mutation, TIL densities within 200 μm of the IM (f_im200) remained the most significant prognostic factor for progression‐free survival (PFS) (hazard ratio [HR] 0.004 [95% confidence interval, CI, 0.0001–0.15], p = 0.0028) in the Yonsei cohort. On multivariate analysis using the TCGA dataset, f_im200 retained prognostic significance for PFS (HR 0.031 [95% CI 0.001–0.645], p = 0.024). Inter‐rater agreement of manual KM grading was insignificant in the Yonsei ( κ = 0.109) and the TCGA ( κ = 0.121) cohorts. The survival analysis based on KM grading showed statistically significant different PFS in the TCGA cohort, but not the Yonsei cohort. Automatic quantification of TILs at the IM based on DL approaches shows prognostic utility to predict PFS, and could provide robust and reproducible TIL density measurement in patients with CRC.