Machine-learning-based integrative -'omics analyses reveal immunologic and metabolic dysregulation in environmental enteric dysfunction.

Environmental enteric dysfunction (EED) is a subclinical enteropathy challenging to diagnose due to an overlap of tissue features with other inflammatory enteropathies. EED subjects (n = 52) from Pakistan, controls (n = 25), and a validation EED cohort (n = 30) from Zambia were used to develop a machine-learning-based image analysis classification model. We extracted histologic feature representations from the Pakistan EED model and correlated them to transcriptomics and clinical biomarkers. In-silico metabolic network modeling was used to characterize alterations in metabolic flux between EED and controls and validated using untargeted lipidomics. Genes encoding beta-ureidopropionase, CYP4F3, and epoxide hydrolase 1 correlated to numerous tissue feature representations. Fatty acid and glycerophospholipid metabolism-related reactions showed altered flux. Increased phosphatidylcholine, lysophosphatidylcholine (LPC), and ether-linked LPCs, and decreased ester-linked LPCs were observed in the duodenal lipidome of Pakistan EED subjects, while plasma levels of glycine-conjugated bile acids were significantly increased. Together, these findings elucidate a multi-omic signature of EED.

The intersection of video capsule endoscopy and artificial intelligence: addressing unique challenges using machine learning.

Introduction: Technical burdens and time-intensive review processes limit the practical utility of video capsule endoscopy (VCE). Artificial intelligence (AI) is poised to address these limitations, but the intersection of AI and VCE reveals challenges that must first be overcome. We identified five challenges to address. Challenge #1: VCE data are stochastic and contains significant artifact. Challenge #2: VCE interpretation is cost-intensive. Challenge #3: VCE data are inherently imbalanced. Challenge #4: Existing VCE AIMLT are computationally cumbersome. Challenge #5: Clinicians are hesitant to accept AIMLT that cannot explain their process. Methods: An anatomic landmark detection model was used to test the application of convolutional neural networks (CNNs) to the task of classifying VCE data. We also created a tool that assists in expert annotation of VCE data. We then created more elaborate models using different approaches including a multi-frame approach, a CNN based on graph representation, and a few-shot approach based on meta-learning. Results: When used on full-length VCE footage, CNNs accurately identified anatomic landmarks (99.1%), with gradient weighted-class activation mapping showing the parts of each frame that the CNN used to make its decision. The graph CNN with weakly supervised learning (accuracy 89.9%, sensitivity of 91.1%), the few-shot model (accuracy 90.8%, precision 91.4%, sensitivity 90.9%), and the multi-frame model (accuracy 97.5%, precision 91.5%, sensitivity 94.8%) performed well. Discussion: Each of these five challenges is addressed, in part, by one of our AI-based models. Our goal of producing high performance using lightweight models that aim to improve clinician confidence was achieved.

Quantitative Morphometry and Machine Learning Model to Explore Duodenal and Rectal Mucosal Tissue of Children with Environmental Enteric Dysfunction.

Environmental enteric dysfunction (EED) is a subclinical enteropathy prevalent in resource-limited settings, hypothesized to be a consequence of chronic exposure to environmental enteropathogens, resulting in malnutrition, growth failure, neurocognitive delays, and oral vaccine failure. This study explored the duodenal and colonic tissues of children with EED, celiac disease, and other enteropathies using quantitative mucosal morphometry, histopathologic scoring indices, and machine learning-based image analysis from archival and prospective cohorts of children from Pakistan and the United States. We observed villus blunting as being more prominent in celiac disease than in EED, as shorter lengths of villi were observed in patients with celiac disease from Pakistan than in those from the United States, with median (interquartile range) lengths of 81 (73, 127) µm and 209 (188, 266) µm, respectively. Additionally, per the Marsh scoring method, celiac disease histologic severity was increased in the cohorts from Pakistan. Goblet cell depletion and increased intraepithelial lymphocytes were features of EED and celiac disease. Interestingly, the rectal tissue from cases with EED showed increased mononuclear inflammatory cells and intraepithelial lymphocytes in the crypts compared with controls. Increased neutrophils in the rectal crypt epithelium were also significantly associated with increased EED histologic severity scores in duodenal tissue. We observed an overlap between diseased and healthy duodenal tissue upon leveraging machine learning image analysis. We conclude that EED comprises a spectrum of inflammation in the duodenum, as previously described, and the rectal mucosa, warranting the examination of both anatomic regions in our efforts to understand and manage EED.

The Data Efficiency of Deep Learning Is Degraded by Unnecessary Input Dimensions.

Biological learning systems are outstanding in their ability to learn from limited training data compared to the most successful learning machines, i.e., Deep Neural Networks (DNNs). What are the key aspects that underlie this data efficiency gap is an unresolved question at the core of biological and artificial intelligence. We hypothesize that one important aspect is that biological systems rely on mechanisms such as foveations in order to reduce unnecessary input dimensions for the task at hand, e.g., background in object recognition, while state-of-the-art DNNs do not. Datasets to train DNNs often contain such unnecessary input dimensions, and these lead to more trainable parameters. Yet, it is not clear whether this affects the DNNs' data efficiency because DNNs are robust to increasing the number of parameters in the hidden layers, and it is uncertain whether this holds true for the input layer. In this paper, we investigate the impact of unnecessary input dimensions on the DNNs data efficiency, namely, the amount of examples needed to achieve certain generalization performance. Our results show that unnecessary input dimensions that are task-unrelated substantially degrade data efficiency. This highlights the need for mechanisms that remove task-unrelated dimensions, such as foveation for image classification, in order to enable data efficiency gains.

A Machine Learning-Based Analysis of Game Data for Attention Deficit Hyperactivity Disorder Assessment.

OBJECTIVE: Attention deficit hyperactivity disorder (ADHD) is found in 9.5 percent of the U.S. population and poses lifelong challenges. Current diagnostic approaches rely on evaluation forms completed by teachers and/or parents, although they are not specifically trained to recognize cognitive disorders. The most accurate diagnosis is by a psychiatrist, often only available to children with severe symptoms. Development of a tool that is engaging and objective and aids medical providers is needed in the diagnosis of ADHD. The goal of this research is to work toward the development of such a tool. MATERIALS AND METHODS: The proposed approach takes advantage of two trends: The rapid adoption of tangible user interface devices and the popularity of interactive videogames. CogCubed Inc. (Minneapolis, MN) has created "Groundskeeper," a game on the Sifteo Cubes (Sifteo, Inc., San Francisco, CA) game system with elements that exercise skills affected by ADHD. "Groundskeeper" was evaluated for 52 patients, with and without ADHD. Gameplay data were mathematically transformed into ADHD-indicative feature variables and subjected to machine learning algorithms to develop diagnostic models to aid psychiatric clinical assessments of ADHD. The effectiveness of the developed model was evaluated against the diagnostic impressions of two licensed child/adolescent psychiatrists using semistructured interviews. RESULTS: Our predictive algorithms were highly accurate in correctly predicting diagnoses based on gameplay of "Groundskeeper." The F-measure, a measure of diagnosis accuracy, from the predictive models gave values as follows: ADHD, inattentive type, 78 percent (P>0.05); ADHD, combined type, 75 percent (P<0.05); anxiety disorders, 71%; and depressive disorders, 76%. CONCLUSIONS: This represents a promising new approach to screening tools for ADHD.