The successes and pitfalls: Deep-learning effectiveness in a Chernobyl field camera trap application.

Camera traps have become in situ sensors for collecting information on animal abundance and occupancy estimates. When deployed over a large landscape, camera traps have become ideal for measuring the health of ecosystems, particularly in unstable habitats where it can be dangerous or even impossible to observe using conventional methods. However, manual processing of imagery is extremely time and labor intensive. Because of the associated expense, many studies have started to employ machine-learning tools, such as convolutional neural networks (CNNs). One drawback for the majority of networks is that a large number of images (millions) are necessary to devise an effective identification or classification model. This study examines specific factors pertinent to camera trap placement in the field that may influence the accuracy metrics of a deep-learning model that has been trained with a small set of images. False negatives and false positives may occur due to a variety of environmental factors that make it difficult for even a human observer to classify, including local weather patterns and daylight. We transfer-trained a CNN to detect 16 different object classes (14 animal species, humans, and fires) across 9576 images taken from camera traps placed in the Chernobyl Exclusion Zone. After analyzing wind speed, cloud cover, temperature, image contrast, and precipitation, there was not a significant correlation between CNN success and ambient conditions. However, a possible positive relationship between temperature and CNN success was noted. Furthermore, we found that the model was more successful when images were taken during the day as well as when precipitation was not present. This study suggests that while qualitative site-specific factors may confuse quantitative classification algorithms such as CNNs, training with a dynamic training set can account for ambient conditions so that they do not have a significant impact on CNN success.

GMP-Compliant Production of Autologous Adipose-Derived Stromal Cells in the NANT 001 Closed Automated Bioreactor.

In recent years mesenchymal stromal cells (MSCs) have received a great deal of interest for the treatment of major diseases, but clinical translation and market authorization have been slow. This has been due in part to a lack of standardization in cell manufacturing protocols, as well as a lack of biologically meaningful cell characterization tools and release assays. Cell production strategies to date have involved complex manual processing in an open environment which is costly, inefficient and poses risks of contamination. The NANT 001 bioreactor has been developed for the automated production of small to medium cell batches for autologous use. This is a closed, benchtop system which automatically performs several processes including cell seeding, media change, real-time monitoring of temperature, pH, cell confluence and cell detachment. Here we describe a validation of the bioreactor in an environment compliant with current good manufacturing practice (cGMP) to confirm its utility in replacing standardized manual processing. Stromal vascular fraction (SVF) was isolated from lipoaspirate material obtained from healthy donors. SVF cells were seeded in the bioreactor. Cell processing was performed automatically and cell harvesting was triggered by computerized analysis of images captured by a travelling microscope positioned beneath the cell culture flask. For comparison, the same protocol was performed in parallel using manual methods. Critical quality attributes (CQA) assessed for cells from each process included cell yield, viability, surface immunophenotype, differentiation propensity, microbial sterility and endotoxin contamination. Cell yields from the bioreactor cultures were comparable in the manual and automated cultures and viability was >90% for both. Expression of surface markers were consistent with standards for adipose-derived stromal cell (ASC) phenotype. ASCs expanded in both automated and manual processes were capable of adipogenic and osteogenic differentiation. Supernatants from all cultures tested negative for microbial and endotoxin contamination. Analysis of labor commitment indicated considerable economic advantage in the automated system in terms of operator, quality control, product release and management personnel. These data demonstrate that the NANT 001 bioreactor represents an effective option for small to medium scale, automated, closed expansion of ASCs from SVF and produces cell products with CQA equivalent to manual processes.

An approach to rapid processing of camera trap images with minimal human input.

Camera traps have become an extensively utilized tool in ecological research, but the manual processing of images created by a network of camera traps rapidly becomes an overwhelming task, even for small camera trap studies.We used transfer learning to create convolutional neural network (CNN) models for identification and classification. By utilizing a small dataset with an average of 275 labeled images per species class, the model was able to distinguish between species and remove false triggers.We trained the model to detect 17 object classes with individual species identification, reaching an accuracy up to 92% and an average F1 score of 85%. Previous studies have suggested the need for thousands of images of each object class to reach results comparable to those achieved by human observers; however, we show that such accuracy can be achieved with fewer images.With transfer learning and an ongoing camera trap study, a deep learning model can be successfully created by a small camera trap study. A generalizable model produced from an unbalanced class set can be utilized to extract trap events that can later be confirmed by human processors.

Corneal Higher-Order Aberrations in Descemet Membrane Endothelial Keratoplasty versus Ultrathin DSAEK in the Descemet Endothelial Thickness Comparison Trial: A Randomized Clinical Trial.

PURPOSE: To compare corneal higher-order aberrations (HOA) after ultrathin Descemet stripping automated endothelial keratoplasty (DSAEK) and Descemet membrane endothelial keratoplasty (DMEK). DESIGN: Patient- and outcome-masked randomized controlled clinical trial. PARTICIPANTS: Patients with damaged or diseased endothelium from Fuchs endothelial dystrophy or pseudophakic bullous keratopathy who were good candidates for DMEK or ultrathin DSAEK. METHODS: Corneal anterior and posterior surface HOA were measured with Scheimpflug imaging before surgery and at 3, 6, and 12 months after surgery. HOA after ultrathin DSAEK and DMEK were compared; correlation was performed between best spectacle-corrected visual acuity (BSCVA) and HOA at each time point. MAIN OUTCOME MEASURES: Higher-order aberrations of the anterior and posterior cornea, expressed as the root mean square deviation from a best fit sphere reference surface. RESULTS: At 3, 6, and 12 months after surgery, the posterior corneal surface had significantly less coma (P ≤ 0.003) and total HOA (P ≤ 0.001) in DMEK compared with ultrathin DSAEK (4.0- and 6.0-mm OZ). Posterior trefoil (P ≤ 0.034), secondary astigmatism (P ≤ 0.042), and tetrafoil (P ≤ 0.045) were lower in DMEK than ultrathin DSAEK at 3, 6, or 12 months (either 4.0- or 6.0-mm OZ). There were no significant differences in anterior surface HOA between DMEK and ultrathin DSAEK at any post-surgical time. Compared with baseline, total posterior HOA was increased (P ≤ 0.036) in ultrathin DSAEK at 3, 6, and 12 months, in contrast to DMEK, where it was decreased (P ≤ 0.044) at 6 and 12 months (4.0- or 6.0-mm OZ, or both). At 6 and 12 months, posterior corneal total HOA correlated with BSCVA (ρ ≤ 0.635, P ≤ 0.001; 4.0- and 6.0-mm OZ). There were no moderate or strong correlations between anterior or combined corneal surface HOA at any time point after surgery. CONCLUSIONS: Descemet membrane endothelial keratoplasty results in less posterior corneal HOA compared with ultrathin DSAEK. Descemet membrane endothelial keratoplasty decreases and ultrathin DSAEK increases posterior corneal HOA compared with presurgical values. Total posterior corneal HOA correlates with 6- and 12-month postoperative visual acuity and may account for the better visual acuity observed after DMEK.

A non-invasive analysis method for on-chip spectrophotometric detection using liquid-core waveguiding within a 3D architecture.

The on-chip measurement of absorbing species has proven to be challenging, particularly with respect to the sample pathlengths available in a miniaturised system. This paper demonstrates how the principles of total internal reflection can be utilised to form a liquid-core waveguide along a single microfluidic channel, increasing the sampling pathlength to 5 mm while maintaining a detection volume of < or = 1 microL. This was achieved using the Teflon fluoropolymers PTFE, FEP and AF as cladding for the liquid-core waveguide. In conjunction with a 3D chip architecture, the use of the liquid-core waveguide enables more efficient use of the probing light beam along with easy and effective coupling of the source, microfluidic chip and the detection system. The confirmation that waveguiding was occurring was successfully demonstrated and the subsequent spectrophotometric analysis of crystal violet provided a linear calibration with reproducibility (< 2.4% RSD) and limits of detection (< 1.3 microM), comparable to absorbance measurements made with a standard UV-Vis spectrophotometer.