Performance of an Artificial Intelligence Model for Recognition and Quantitation of Histologic Features of Eosinophilic Esophagitis on Biopsy Samples.

We have developed an artificial intelligence (AI)-based digital pathology model for the evaluation of histologic features related to eosinophilic esophagitis (EoE). In this study, we evaluated the performance of our AI model in a cohort of pediatric and adult patients for histologic features included in the Eosinophilic Esophagitis Histologic Scoring System (EoEHSS). We collected a total of 203 esophageal biopsy samples from patients with mucosal eosinophilia of any degree (91 adult and 112 pediatric patients) and 10 normal controls from a prospectively maintained database. All cases were assessed by a specialized gastrointestinal (GI) pathologist for features in the EoEHSS at the time of original diagnosis and rescored by a central GI pathologist (R.K.M.). We subsequently analyzed whole-slide image digital slides using a supervised AI model operating in a cloud-based, deep learning AI platform (Aiforia Technologies) for peak eosinophil count (PEC) and several histopathologic features in the EoEHSS. The correlation and interobserver agreement between the AI model and pathologists (Pearson correlation coefficient [rs] = 0.89 and intraclass correlation coefficient [ICC] = 0.87 vs original pathologist; rs = 0.91 and ICC = 0.83 vs central pathologist) were similar to the correlation and interobserver agreement between pathologists for PEC (rs = 0.88 and ICC = 0.91) and broadly similar to those for most other histologic features in the EoEHSS. The AI model also accurately identified PEC of >15 eosinophils/high-power field by the original pathologist (area under the curve [AUC] = 0.98) and central pathologist (AUC = 0.98) and had similar AUCs for the presence of EoE-related endoscopic features to pathologists' assessment. Average eosinophils per epithelial unit area had similar performance compared to AI high-power field-based analysis. Our newly developed AI model can accurately identify, quantify, and score several of the main histopathologic features in the EoE spectrum, with agreement regarding EoEHSS scoring which was similar to that seen among GI pathologists.

Development and technical validation of an artificial intelligence model for quantitative analysis of histopathologic features of eosinophilic esophagitis.

Background: In an attempt to provide quantitative, reproducible, and standardized analyses in cases of eosinophilic esophagitis (EoE), we have developed an artificial intelligence (AI) digital pathology model for the evaluation of histologic features in the EoE/esophageal eosinophilia spectrum. Here, we describe the development and technical validation of this novel AI tool. Methods: A total of 10 726 objects and 56.2 mm2 of semantic segmentation areas were annotated on whole-slide images, utilizing a cloud-based, deep learning artificial intelligence platform (Aiforia Technologies, Helsinki, Finland). Our training set consisted of 40 carefully selected digitized esophageal biopsy slides which contained the full spectrum of changes typically seen in the setting of esophageal eosinophilia, ranging from normal mucosa to severe abnormalities with regard to each specific features included in our model. A subset of cases was reserved as independent "test sets" in order to assess the validity of the AI model outside the training set. Five specialized experienced gastrointestinal pathologists scored each feature blindly and independently of each other and of AI model results. Results: The performance of the AI model for all cell type features was similar/non-inferior to that of our group of GI pathologists (F1-scores: 94.5-94.8 for AI vs human and 92.6-96.0 for human vs human). Segmentation area features were rated for accuracy using the following scale: 1. "perfect or nearly perfect" (95%-100%, no significant errors), 2. "very good" (80%-95%, only minor errors), 3. "good" (70%-80%, significant errors but still captures the feature well), 4. "insufficient" (less than 70%, significant errors compromising feature recognition). Rating scores for tissue (1.01), spongiosis (1.15), basal layer (1.05), surface layer (1.04), lamina propria (1.15), and collagen (1.11) were in the "very good" to "perfect or nearly perfect" range, while degranulation (2.23) was rated between "good" and "very good". Conclusion: Our newly developed AI-based tool showed an excellent performance (non-inferior to a group of experienced GI pathologists) for the recognition of various histologic features in the EoE/esophageal mucosal eosinophilia spectrum. This tool represents an important step in creating an accurate and reproducible method for semi-automated quantitative analysis to be used in the evaluation of esophageal biopsies in this clinical context.

Artificial intelligence identifies inflammation and confirms fibroblast foci as prognostic tissue biomarkers in idiopathic pulmonary fibrosis.

A large number of fibroblast foci (FF) predict mortality in idiopathic pulmonary fibrosis (IPF). Other prognostic histological markers have not been identified. Artificial intelligence (AI) offers a possibility to quantitate possible prognostic histological features in IPF. We aimed to test the use of AI in IPF lung tissue samples by quantitating FF, interstitial mononuclear inflammation, and intra-alveolar macrophages with a deep convolutional neural network (CNN). Lung tissue samples of 71 patients with IPF from the FinnishIPF registry were analyzed by an AI model developed in the Aiforia® platform. The model was trained to detect tissue, air spaces, FF, interstitial mononuclear inflammation, and intra-alveolar macrophages with 20 samples. For survival analysis, cut-point values for high and low values of histological parameters were determined with maximally selected rank statistics. Survival was analyzed using the Kaplan-Meier method. A large area of FF predicted poor prognosis in IPF (p = 0.01). High numbers of interstitial mononuclear inflammatory cells and intra-alveolar macrophages were associated with prolonged survival (p = 0.01 and p = 0.01, respectively). Of lung function values, low diffusing capacity for carbon monoxide was connected to a high density of FF (p = 0.03) and a high forced vital capacity of predicted was associated with a high intra-alveolar macrophage density (p = 0.03). The deep CNN detected histological features that are difficult to quantitate manually. Interstitial mononuclear inflammation and intra-alveolar macrophages were novel prognostic histological biomarkers in IPF. Evaluating histological features with AI provides novel information on the prognostic estimation of IPF.