ABSTRACT
Emerging deep learning-based applications in precision medicine include computational histopathological analysis. However, there is a lack of the required training image datasets to generate classification and detection models. This phenomenon occurs mainly due to human factors that make it difficult to obtain well-annotated data. The present study provides a curated public collection of histopathological images (DeepHP) and a convolutional neural network model for diagnosing gastritis. Images from gastric biopsy histopathological exams were used to investigate the performance of the proposed model in detecting gastric mucosa with Helicobacter pylori infection. The DeepHP database comprises 394,926 histopathological images, of which 111 K were labeled as Helicobacter pylori positive and 283 K were Helicobacter pylori negative. We investigated the classification performance of three Convolutional Neural Network architectures. The models were tested and validated with two distinct image sets of 15% (59K patches) chosen randomly. The VGG16 architecture showed the best results with an Area Under the Curve of 0.998%. The results showed that CNN could be used to classify histopathological images from gastric mucosa with marked precision. Our model evidenced high potential and application in the computational pathology field.
Subject(s)
Gastritis , Helicobacter Infections , Helicobacter pylori , Humans , Helicobacter Infections/diagnosis , Helicobacter Infections/pathology , Gastric Mucosa/pathology , Gastritis/diagnosis , Gastritis/pathology , Gastroscopy/methodsABSTRACT
ClinVar is a web platform that stores â¼789,000 genetic associations with complex diseases. A partial set of these cataloged genetic associations has challenged clinicians and geneticists, often leading to conflicting interpretations or uncertain clinical impact significance. In this study, we addressed the (re)classification of genetic variants by AmazonForest, which is a random-forest-based pathogenicity metaprediction model that works by combining functional impact data from eight prediction tools. We evaluated the performance of representation learning algorithms such as autoencoders to propose a better strategy. All metaprediction models were trained with ClinVar data, and genetic variants were annotated with eight functional impact predictors cataloged with SnpEff/SnpSift. AmazonForest implements the best random forest model with a one hot data-encoding strategy, which shows an Area Under ROC Curve of ≥0.93. AmazonForest was employed for pathogenicity prediction of a set of â¼101,000 genetic variants of uncertain significance or conflict of interpretation. Our findings revealed â¼24,000 variants with high pathogenic probability (RFprob≥0.9). In addition, we show results for Alzheimer's Disease as a demonstration of its application in clinical interpretation of genetic variants in complex diseases. Lastly, AmazonForest is available as a web tool and R object that can be loaded to perform pathogenicity predictions.