Automated Identification of Incomplete and Complete Retinal Epithelial Pigment and Outer Retinal Atrophy Using Machine Learning.

OBJECTIVE: To assess and validate a deep learning algorithm to automatically detect incomplete retinal pigment epithelial and outer retinal atrophy (iRORA) and complete retinal pigment epithelial and outer retinal atrophy (cRORA) in eyes with age-related macular degeneration. DESIGN: In a retrospective machine learning analysis, a deep learning model was trained to jointly classify the presence of iRORA and cRORA within a given B-scan. The algorithm was evaluated using 2 separate and independent datasets. PARTICIPANTS: OCT B-scan volumes from 71 patients with nonneovascular age-related macular degeneration captured at the Doheny-University of California Los Angeles Eye Centers and the following 2 external OCT B-scans testing datasets: (1) University of Pennsylvania, University of Miami, and Case Western Reserve University and (2) Doheny Image Reading Research Laboratory. METHODS: The images were annotated by an experienced grader for the presence of iRORA and cRORA. A Resnet18 model was trained to classify these annotations for each B-scan using OCT volumes collected at the Doheny-University of California Los Angeles Eye Centers. The model was applied to 2 testing datasets to assess out-of-sample model performance. MAIN OUTCOMES MEASURES: Model performance was quantified in terms of area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC). Sensitivity, specificity, and positive predictive value were also compared against additional clinician annotators. RESULTS: On an independently collected test set, consisting of 1117 volumes from the general population, the model predicted iRORA and cRORA presence within the entire volume with nearly perfect AUROC performance and AUPRC scores (iRORA, 0.61; 95% confidence interval [CI] [0.45, 0.82]: cRORA, 0.83; 95% CI [0.68, 0.95]). On another independently collected set, consisting of 60 OCT B-scans enriched for iRORA and cRORA lesions, the model performed with AUROC (iRORA: 0.68, 95% CI [0.54, 0.81]; cRORA: 0.84, 95% CI [0.75, 0.94]) and AUPRC (iRORA: 0.70, 95% CI [0.55, 0.86]; cRORA: 0.82, 95% CI [0.70, 0.93]). CONCLUSIONS: A deep learning model can accurately and precisely identify both iRORA and cRORA lesions within the OCT B-scan volume. The model can achieve similar sensitivity compared with human graders, which potentially obviates a laborious and time-consuming annotation process and could be developed into a diagnostic screening tool.

Deep learning-based phenotype imputation on population-scale biobank data increases genetic discoveries.

Biobanks that collect deep phenotypic and genomic data across many individuals have emerged as a key resource in human genetics. However, phenotypes in biobanks are often missing across many individuals, limiting their utility. We propose AutoComplete, a deep learning-based imputation method to impute or 'fill-in' missing phenotypes in population-scale biobank datasets. When applied to collections of phenotypes measured across ~300,000 individuals from the UK Biobank, AutoComplete substantially improved imputation accuracy over existing methods. On three traits with notable amounts of missingness, we show that AutoComplete yields imputed phenotypes that are genetically similar to the originally observed phenotypes while increasing the effective sample size by about twofold on average. Further, genome-wide association analyses on the resulting imputed phenotypes led to a substantial increase in the number of associated loci. Our results demonstrate the utility of deep learning-based phenotype imputation to increase power for genetic discoveries in existing biobank datasets.

Phenotype integration improves power and preserves specificity in biobank-based genetic studies of major depressive disorder.

Biobanks often contain several phenotypes relevant to diseases such as major depressive disorder (MDD), with partly distinct genetic architectures. Researchers face complex tradeoffs between shallow (large sample size, low specificity/sensitivity) and deep (small sample size, high specificity/sensitivity) phenotypes, and the optimal choices are often unclear. Here we propose to integrate these phenotypes to combine the benefits of each. We use phenotype imputation to integrate information across hundreds of MDD-relevant phenotypes, which significantly increases genome-wide association study (GWAS) power and polygenic risk score (PRS) prediction accuracy of the deepest available MDD phenotype in UK Biobank, LifetimeMDD. We demonstrate that imputation preserves specificity in its genetic architecture using a novel PRS-based pleiotropy metric. We further find that integration via summary statistics also enhances GWAS power and PRS predictions, but can introduce nonspecific genetic effects depending on input. Our work provides a simple and scalable approach to improve genetic studies in large biobanks by integrating shallow and deep phenotypes.

SLIViT: a general AI framework for clinical-feature diagnosis from limited 3D biomedical-imaging data.

We present SLIViT, a deep-learning framework that accurately measures disease-related risk factors in volumetric biomedical imaging, such as magnetic resonance imaging (MRI) scans, optical coherence tomography (OCT) scans, and ultrasound videos. To evaluate SLIViT, we applied it to five different datasets of these three different data modalities tackling seven learning tasks (including both classification and regression) and found that it consistently and significantly outperforms domain-specific state-of-the-art models, typically improving performance (ROC AUC or correlation) by 0.1-0.4. Notably, compared to existing approaches, SLIViT can be applied even when only a small number of annotated training samples is available, which is often a constraint in medical applications. When trained on less than 700 annotated volumes, SLIViT obtained accuracy comparable to trained clinical specialists while reducing annotation time by a factor of 5,000 demonstrating its utility to automate and expedite ongoing research and other practical clinical scenarios.

STENSL: Microbial Source Tracking with ENvironment SeLection.

Microbial source tracking analysis has emerged as a widespread technique for characterizing the properties of complex microbial communities. However, this analysis is currently limited to source environments sampled in a specific study. In order to expand the scope beyond one single study and allow the exploration of source environments using large databases and repositories, such as the Earth Microbiome Project, a source selection procedure is required. Such a procedure will allow differentiating between contributing environments and nuisance ones when the number of potential sources considered is high. Here, we introduce STENSL (microbial Source Tracking with ENvironment SeLection), a machine learning method that extends common microbial source tracking analysis by performing an unsupervised source selection and enabling sparse identification of latent source environments. By incorporating sparsity into the estimation of potential source environments, STENSL improves the accuracy of true source contribution, while significantly reducing the noise introduced by noncontributing ones. We therefore anticipate that source selection will augment microbial source tracking analyses, enabling exploration of multiple source environments from publicly available repositories while maintaining high accuracy of the statistical inference. IMPORTANCE Microbial source tracking is a powerful tool to characterize the properties of complex microbial communities. However, this analysis is currently limited to source environments sampled in a specific study. In many applications there is a clear need to consider source selection over a large array of microbial environments, external to the study. To this end, we developed STENSL (microbial Source Tracking with ENvironment SeLection), an expectation-maximization algorithm with sparsity that enables the identification of contributing sources among a large set of potential microbial environments. With the unprecedented expansion of microbiome data repositories such as the Earth Microbiome Project, recording over 200,000 samples from more than 50 types of categorized environments, STENSL takes the first steps in performing automated source exploration and selection. STENSL is significantly more accurate in identifying the contributing sources as well as the unknown source, even when considering hundreds of potential source environments, settings in which state-of-the-art microbial source tracking methods add considerable error.

High Precision Mammography Lesion Identification From Imprecise Medical Annotations.

Breast cancer screening using Mammography serves as the earliest defense against breast cancer, revealing anomalous tissue years before it can be detected through physical screening. Despite the use of high resolution radiography, the presence of densely overlapping patterns challenges the consistency of human-driven diagnosis and drives interest in leveraging state-of-art localization ability of deep convolutional neural networks (DCNN). The growing availability of digitized clinical archives enables the training of deep segmentation models, but training using the most widely available form of coarse hand-drawn annotations works against learning the precise boundary of cancerous tissue in evaluation, while producing results that are more aligned with the annotations rather than the underlying lesions. The expense of collecting high quality pixel-level data in the field of medical science makes this even more difficult. To surmount this fundamental challenge, we propose LatentCADx, a deep learning segmentation model capable of precisely annotating cancer lesions underlying hand-drawn annotations, which we procedurally obtain using joint classification training and a strict segmentation penalty. We demonstrate the capability of LatentCADx on a publicly available dataset of 2,620 Mammogram case files, where LatentCADx obtains classification ROC of 0.97, AP of 0.87, and segmentation AP of 0.75 (IOU = 0.5), giving comparable or better performance than other models. Qualitative and precision evaluation of LatentCADx annotations on validation samples reveals that LatentCADx increases the specificity of segmentations beyond that of existing models trained on hand-drawn annotations, with pixel level specificity reaching a staggering value of 0.90. It also obtains sharp boundary around lesions unlike other methods, reducing the confused pixels in the output by more than 60%.

A machine learning algorithm to increase COVID-19 inpatient diagnostic capacity.

Worldwide, testing capacity for SARS-CoV-2 is limited and bottlenecks in the scale up of polymerase chain reaction (PCR-based testing exist. Our aim was to develop and evaluate a machine learning algorithm to diagnose COVID-19 in the inpatient setting. The algorithm was based on basic demographic and laboratory features to serve as a screening tool at hospitals where testing is scarce or unavailable. We used retrospectively collected data from the UCLA Health System in Los Angeles, California. We included all emergency room or inpatient cases receiving SARS-CoV-2 PCR testing who also had a set of ancillary laboratory features (n = 1,455) between 1 March 2020 and 24 May 2020. We tested seven machine learning models and used a combination of those models for the final diagnostic classification. In the test set (n = 392), our combined model had an area under the receiver operator curve of 0.91 (95% confidence interval 0.87-0.96). The model achieved a sensitivity of 0.93 (95% CI 0.85-0.98), specificity of 0.64 (95% CI 0.58-0.69). We found that our machine learning algorithm had excellent diagnostic metrics compared to SARS-CoV-2 PCR. This ensemble machine learning algorithm to diagnose COVID-19 has the potential to be used as a screening tool in hospital settings where PCR testing is scarce or unavailable.