Objectives-To develop and validate a proof-of-concept convolutional neural network (CNN)based deep learning system (DLS) that classifies common hepatic lesions on multi-phasic MRI.Methods-A custom CNN was engineered by iteratively optimizing the network architecture and training cases, finally consisting of three convolutional layers with associated rectified linear units, two maximum pooling layers, and two fully connected layers. Four hundred ninety-four hepatic lesions with typical imaging features from six categories were utilized, divided into training (n = 434) and test (n = 60) sets. Established augmentation techniques were used to generate 43,400 training samples. An Adam optimizer was used for training. Monte Carlo cross-validation was performed. After model engineering was finalized, classification accuracy for the final CNN was compared with two board-certified radiologists on an identical unseen test set.

BackgroundCells of the oligodendrocyte (OL) lineage play a vital role in the production and maintenance of myelin, a multilamellar membrane which allows for saltatory conduction along axons. These cells may provide immense therapeutic potential for lost sensory and motor function in demyelinating conditions, such as spinal cord injury, multiple sclerosis, and transverse myelitis. However, the molecular mechanisms controlling OL differentiation are largely unknown. MicroRNAs (miRNAs) are considered the “micromanagers” of gene expression with suggestive roles in cellular differentiation and maintenance. Although unique patterns of miRNA expression in various cell lineages have been characterized, this is the first report documenting their expression during oligodendrocyte maturation from human embryonic stem (hES) cells. Here, we performed a global miRNA analysis to reveal and identify characteristic patterns in the multiple stages leading to OL maturation from hES cells including those targeting factors involved in myelin production.Methodology/Principal FindingsWe isolated cells from 8 stages of OL differentiation. Total RNA was subjected to miRNA profiling and validations preformed using real-time qRT-PCR. A comparison of miRNAs from our cultured OLs and OL progenitors showed significant similarities with published results from equivalent cells found in the rat and mouse central nervous system. Principal component analysis revealed four main clusters of miRNA expression corresponding to early, mid, and late progenitors, and mature OLs. These results were supported by correlation analyses between adjacent stages. Interestingly, the highest differentially-expressed miRNAs demonstrated a similar pattern of expression throughout all stages of differentiation, suggesting that they potentially regulate a common target or set of targets in this process. The predicted targets of these miRNAs include those with known or suspected roles in oligodendrocyte development and myelination including C11Orf9, CLDN11, MYTL1, MBOP, MPZL2, and DDR1.Conclusions/SignificanceWe demonstrate miRNA profiles during distinct stages in oligodendroglial differentiation that may provide key markers of OL maturation. Our results reveal pronounced trends in miRNA expression and their potential mRNA target interactions that could provide valuable insight into the molecular mechanisms of differentiation.

Objectives-To develop a proof-of-concept "interpretable" deep learning prototype that justifies aspects of its predictions from a pre-trained hepatic lesion classifier. Methods-A convolutional neural network (CNN) was engineered and trained to classify six hepatic tumor entities using 494 lesions on multi-phasic MRI, described in Part 1. A subset of each lesion class was labeled with up to four key imaging features per lesion. A post hoc algorithm inferred the presence of these features in a test set of 60 lesions by analyzing activation patterns of the pre-trained CNN model. Feature maps were generated that highlight regions in the original image that correspond to particular features. Additionally, relevance scores were assigned to each identified feature, denoting the relative contribution of a feature to the predicted lesion classification.

This study utilized a contusion model of spinal cord injury (SCI) in rats using the standardized NYU-MASCIS impactor, after which oligodendrocyte progenitor cells (OPCs) derived from human embryonic stem cell (ESC) were transplanted into the spinal cord to study their survival and migration route toward the areas of injury. One critical aspect of successful cell-based SCI therapy is the time of injection following injury. OPCs were injected at two clinically relevant times when most damage occurs to the surrounding tissue, 3 and 24 hours following injury. Migration and survivability after eight days was measured postmortem. In-vitro immunofluorescence revealed that most ESC-derived OPCs expressed oligodendrocyte markers, including CNPase, GalC, Olig1, O4, and O1. Results showed that OPCs survived when injected at the center of injury and migrated away from the injection sites after one week. Histological sections revealed integration of ESC-derived OPCs into the spinal cord with contusion injury without disruption to the parenchyma. Cells survived for a minimum of eight days after injury, without tumor or cyst formation. The extent of injury and effect of early cell transplant was measured using behavioral and electrophysiological assessments which demonstrated increased neurological responses in rats transplanted with OPCs compared to controls.