Identification of pancreatic adenocarcinoma molecular subtypes on histology slides using deep learning models
Abstract
Background
Pancreatic adenocarcinoma (PAC) is predicted to be the second cause of death by cancer in 2030 and its prognosis has seen little improvement in the last decades. PAC is a very heterogeneous tumor with preeminent stroma and multiple histological aspects. Omic studies confirmed its molecular heterogeneity, possibly one of the main factors explaining the failure of most clinical trials. Two and three transcriptomic subtypes of tumor cells and stroma respectively, were described with major prognostic and predictive implications. The tumor subtypes, Basal-like and Classical, have been shown by several groups to be predictive of the response to first line chemotherapy. As of today, these subtypes can only be defined by RNA profiling which is limited by the quantity and quality of the samples (formalin fixation and low cellularity) as well as by the analytical delay that may restrict its application in routine care. In addition, tumors may harbor a mixture of several subtypes limiting their interpretation using bulk transcriptomic approaches and thereby their clinical use. Here, we propose a multistep approach using deep learning models to predict tumor components and their molecular subtypes on routine histological preparations.
Methods
728 whole-slide digitized histological slides corresponding to 350 consecutive resected PAC from four centers with clinical and transcriptomic data were assembled and used as a discovery set. PAC from TCGA (n = 134) was used as a validation set. Tumor regions from slides of the discovery set were annotated to train a multistep deep learning model that first recognizes tumor tissue and then predicts tumor and stroma cells molecular subtypes assessed by the published PurIST algorithm.
Results
The tumor detection model was very efficient (AUC = 0.98 in the TCGA validation cohort). In the discovery set, the Basal-like/Classical classification performance of the model by cross validation was 0.79 (AUC) and reached 0.86 when restricted to samples with a high-confidence RNA-defined molecular subtype.Subtypes defined by the model were independently associated with overall survival in multivariate analysis (HR = 2.56 [1.87 - 3.49], pval < 0.001), and association was higher relatively to PurIST RNA subtypes (HR = 1.60 [1.17 - 2.19] pval < 0.001). In the validation cohort, the model had an overall AUC of 0.82, and 0.89 in the subset of “subtype-pure” tumors. In addition to demonstrating the value of histology-based deep learning models for tumor subtyping in PAC, these results also show the limit of molecular-based subtyping in highly heterogeneous samples.
Conclusions
This study provides the first PAC subtyping tool usable worldwide in clinical practice, finally opening the possibility of patient molecular stratification in routine care and clinical trials.
