Human selection bias drives the linear nature of the more ground truth effect in explainable deep learning optical coherence tomography image segmentation.

Supervised deep learning (DL) algorithms are highly dependent on training data for which human graders are assigned, for example, for optical coherence tomography (OCT) image annotation. Despite the tremendous success of DL, due to human judgment, these ground truth labels can be inaccurate and/or ambiguous and cause a human selection bias. We therefore investigated the impact of the size of the ground truth and variable numbers of graders on the predictive performance of the same DL architecture and repeated each experiment three times. The largest training dataset delivered a prediction performance close to that of human experts. All DL systems utilized were highly consistent. Nevertheless, the DL under-performers could not achieve any further autonomous improvement even after repeated training. Furthermore, a quantifiable linear relationship between ground truth ambiguity and the beneficial effect of having a larger amount of ground truth data was detected and marked as the more-ground-truth effect.

Cynomolgus monkey's retina volume reference database based on hybrid deep learning optical coherence tomography segmentation.

Cynomolgus monkeys (Macaca fascicularis) are commonly used in pre-clinical ocular studies. However, studies that report the morphological features of the macaque retina are based only on minimal sample sizes; therefore, little is known about the normal distribution and background variation. This study was conducted using optical coherence tomography (OCT) imaging to investigate the variations in retinal volumes of healthy cynomolgus monkeys and the effects of sex, origin, and eye side on the retinal volumes to establish a comprehensive reference database. A machine-learning algorithm was employed to segment the retina within the OCT data (i.e., generated pixel-wise labels). Furthermore, a classical computer vision algorithm has identified the deepest point in a foveolar depression. The retinal volumes were determined and analyzed based on this reference point and segmented retinal compartments. Notably, the overall foveolar mean volume in zone 1, which is the region of the sharpest vision, was 0.205 mm3 (range 0.154-0.268 mm3), with a relatively low coefficient of variation of 7.9%. Generally, retinal volumes exhibit a relatively low degree of variation. However, significant differences in the retinal volumes due to the monkey's origin were identified. Additionally, sex had a significant impact on the paracentral retinal volume. Therefore, the origin and sex of cynomolgus monkeys should be considered when evaluating the macaque retinal volumes based on this dataset.

Iris Color Matters-A Contractility Analysis With Dynamic Volume-Rendered Optical Coherence Tomography Pupillometry.

Purpose: To analyze natural variability in pupillary contractility with dynamic volume-rendered optical coherence tomography (OCT) pupillometry regarding iris color, age, and sex in healthy Caucasian participants. Methods: The intrapupillary spaces (IPSs) derived from anterior segment swept-source OCT of 71 healthy eyes were retrospectively analyzed. Baseline scotopic and photopic volumes and the functional parameters of pupillary ejection fraction (PEF), three-dimensional (3D) contractility, and relative light response (RLR) were measured on the swept-source OCT volumes. The effect on these parameters of iris color (brown, green, and blue), age, and sex was assessed. Results: More pigmented irises were more contractile than less pigmented irises. Iris color significantly affected scotopic baseline IPSs (brown, 10.39 ± 4.86 mm3; green, 9.68 ± 3.31 mm3; blue, 6.75 ± 4.27 mm3; P = 0.018), PEF (brown, 90.8% ± 2.7%; green, 89.1% ± 2.5%; blue, 85.0% ± 9.3%; P = 0.010), 3D contractility (brown, 9.52 ± 4.59 mm3; green, 8.66 ± 3.07 mm3; blue, 6.44 ± 4.87 mm3; P = 0.016), and RLR (brown, 11.90 ± 4.03; green, 9.75 ± 2.73; blue, 8.52 ± 3.88; P = 0.026). Absolute scotopic volume (P = 0.022) and 3D contractility (P = 0.024) decreased with age. Sex showed no correlations. Conclusions: The natural variability of pupillary contractility can be analyzed with dynamic OCT pupillometry. Iris color and age can impact pupillary response with this method. Translational Relevance: Iris contractility parameters can be measured using a commercially available OCT system, allowing for quantification of the aqueous humor volume inside the pupil.

Feasibility of Automated Segmentation of Pigmented Choroidal Lesions in OCT Data With Deep Learning.

Purpose: To evaluate the feasibility of automated segmentation of pigmented choroidal lesions (PCLs) in optical coherence tomography (OCT) data and compare the performance of different deep neural networks. Methods: Swept-source OCT image volumes were annotated pixel-wise for PCLs and background. Three deep neural network architectures were applied to the data: the multi-dimensional gated recurrent units (MD-GRU), the V-Net, and the nnU-Net. The nnU-Net was used to compare the performance of two-dimensional (2D) versus three-dimensional (3D) predictions. Results: A total of 121 OCT volumes were analyzed (100 normal and 21 PCLs). Automated PCL segmentations were successful with all neural networks. The 3D nnU-Net predictions showed the highest recall with a mean of 0.77 ± 0.22 (MD-GRU, 0.60 ± 0.31; V-Net, 0.61 ± 0.25). The 3D nnU-Net predicted PCLs with a Dice coefficient of 0.78 ± 0.13, outperforming MD-GRU (0.62 ± 0.23) and V-Net (0.59 ± 0.24). The smallest distance to the manual annotation was found using 3D nnU-Net with a mean maximum Hausdorff distance of 315 ± 172 µm (MD-GRU, 1542 ± 1169 µm; V-Net, 2408 ± 1060 µm). The 3D nnU-Net showed a superior performance compared with stacked 2D predictions. Conclusions: The feasibility of automated deep learning segmentation of PCLs was demonstrated in OCT data. The neural network architecture had a relevant impact on PCL predictions. Translational Relevance: This work serves as proof of concept for segmentations of choroidal pathologies in volumetric OCT data; improvements are conceivable to meet clinical demands for the diagnosis, monitoring, and treatment evaluation of PCLs.

Graphene-Derived Carbon Support Boosts Proton Exchange Membrane Fuel Cell Catalyst Stability.

The lack of efficient and durable proton exchange membrane fuel cell electrocatalysts for the oxygen reduction reaction is still restraining the present hydrogen technology. Graphene-based carbon materials have emerged as a potential solution to replace the existing carbon black (CB) supports; however, their potential was never fully exploited as a commercial solution because of their more demanding properties. Here, a unique and industrially scalable synthesis of platinum-based electrocatalysts on graphene derivative (GD) supports is presented. With an innovative approach, highly homogeneous as well as high metal loaded platinum-alloy (up to 60 wt %) intermetallic catalysts on GDs are achieved. Accelerated degradation tests show enhanced durability when compared to the CB-supported analogues including the commercial benchmark. Additionally, in combination with X-ray photoelectron spectroscopy Auger characterization and Raman spectroscopy, a clear connection between the sp 2 content and structural defects in carbon materials with the catalyst durability is observed. Advanced gas diffusion electrode results show that the GD-supported catalysts exhibit excellent mass activities and possess the properties necessary to reach high currents if utilized correctly. We show record-high peak power densities in comparison to the prior best literature on platinum-based GD-supported materials which is promising information for future application.

Accuracy of Trained Physicians is Inferior to Deep Learning-Based Algorithm for Determining Angles in Ultrasound of the Newborn Hip.

PURPOSE: Sonographic diagnosis of developmental dysplasia of the hip allows treatment with a flexion-abduction orthosis preventing hip luxation. Accurate determination of alpha and beta angles according to Graf is crucial for correct diagnosis. It is unclear if algorithms could predict the angles. We aimed to compare the accuracy for users and automation reporting root mean squared errors (RMSE). MATERIALS AND METHODS: We used 303 306 ultrasound images of newborn hips collected between 2009 and 2016 in screening consultations. Trained physicians labelled every second image with alpha and beta angles during the consultations. A random subset of images was labeled with time and precision under lab conditions as ground truth. Automation predicted the two angles using a convolutional neural network (CNN). The analysis was focused on the alpha angle. RESULTS: Three methods were implemented, each with a different abstraction of the problem: (1) CNNs that directly learn the angles without any post-processing steps; (2) CNNs that return the relevant landmarks in the image to identify the angles; (3) CNNs that return the base line, bony roof line, and the cartilage roof line which are necessary to calculate the angles. The RMSE between physicians and ground truth were found to be 7.1° for alpha. The best CNN architecture was (2) landmark detection. The RMSE between landmark detection and ground truth was 3.9° for alpha. CONCLUSION: The accuracy of physicians in their daily routine is inferior to deep learning-based algorithms for determining angles in ultrasound of the newborn hip. Similar methods could be used to support physicians.

Unraveling the deep learning gearbox in optical coherence tomography image segmentation towards explainable artificial intelligence.

Machine learning has greatly facilitated the analysis of medical data, while the internal operations usually remain intransparent. To better comprehend these opaque procedures, a convolutional neural network for optical coherence tomography image segmentation was enhanced with a Traceable Relevance Explainability (T-REX) technique. The proposed application was based on three components: ground truth generation by multiple graders, calculation of Hamming distances among graders and the machine learning algorithm, as well as a smart data visualization ('neural recording'). An overall average variability of 1.75% between the human graders and the algorithm was found, slightly minor to 2.02% among human graders. The ambiguity in ground truth had noteworthy impact on machine learning results, which could be visualized. The convolutional neural network balanced between graders and allowed for modifiable predictions dependent on the compartment. Using the proposed T-REX setup, machine learning processes could be rendered more transparent and understandable, possibly leading to optimized applications.

Feasibility and Safety of a Coaxial Dual-Wavelength Optical Coherence Tomography Apparatus.

PURPOSE: To evaluate the feasibility and safety of a coaxial dual-wavelength optical coherence tomography (OCT) device (marked as Hydra-OCT). METHODS: Healthy participants without ocular pathology underwent retinal imaging using the Hydra-OCT allowing for simultaneous measurement of retinal scanning of 840 and 1,072 nm wavelength. Before and after measurement, best-corrected visual acuity and patients' comfort were assessed. Representative OCT images from both wavelengths were compared by 5 independent graders using a subjective grading scheme. RESULTS: A total of 30 eyes of 30 participants (8 females and 22 males) with a mean age of 26.5 years (range from 19 to 55 years) were included. Dual-wavelength image acquisition was made possible in each subject. The participant's effort and comfort assessment using the Hydra-OCT imaging revealed an equivalent value as compared to the commercially available OCT machine. No adverse events were reported, and visual acuity was not altered by the Hydra-OCT. Imaging between the systems was comparable. CONCLUSIONS: This study provides evidence for the feasibility and safety of a coaxial dual-wavelength OCT imaging method under real-life conditions. The novel Hydra-OCT imaging device may offer additional insights into the pathology of retinal and choroidal diseases.

Toward the Continuous Production of Multigram Quantities of Highly Uniform Supported Metallic Nanoparticles and Their Application for Synthesis of Superior Intermetallic Pt-Alloy ORR Electrocatalysts.

A fast and facile pulse combustion (PC) method that allows for the continuous production of multigram quantities of high-metal-loaded and highly uniform supported metallic nanoparticles (SMNPs) is presented. Namely, various metal on carbon (M/C) composites have been prepared by using only three feedstock components: water, metal-salt, and the supporting material. The present approach can be elegantly utilized also for numerous other applications in electrocatalysis, heterogeneous catalysis, and sensors. In this study, the PC-prepared M/C composites were used as metal precursors for the Pt NPs deposition using double passivation with the galvanic displacement method (DP method). Lastly, by using thin-film rotating disc electrode (TF-RDE) and gas-diffusion electrode (GDE) methodologies, we show that the synergistic effects of combining PC technology with the DP method enable production of superior intermetallic Pt-M electrocatalysts with an improved oxygen reduction reaction (ORR) performance when compared to a commercial Pt-Co electrocatalyst for proton exchange membrane fuel cells (PEMFCs) application.

Validation of automated artificial intelligence segmentation of optical coherence tomography images.

PURPOSE: To benchmark the human and machine performance of spectral-domain (SD) and swept-source (SS) optical coherence tomography (OCT) image segmentation, i.e., pixel-wise classification, for the compartments vitreous, retina, choroid, sclera. METHODS: A convolutional neural network (CNN) was trained on OCT B-scan images annotated by a senior ground truth expert retina specialist to segment the posterior eye compartments. Independent benchmark data sets (30 SDOCT and 30 SSOCT) were manually segmented by three classes of graders with varying levels of ophthalmic proficiencies. Nine graders contributed to benchmark an additional 60 images in three consecutive runs. Inter-human and intra-human class agreement was measured and compared to the CNN results. RESULTS: The CNN training data consisted of a total of 6210 manually segmented images derived from 2070 B-scans (1046 SDOCT and 1024 SSOCT; 630 C-Scans). The CNN segmentation revealed a high agreement with all grader groups. For all compartments and groups, the mean Intersection over Union (IOU) score of CNN compartmentalization versus group graders' compartmentalization was higher than the mean score for intra-grader group comparison. CONCLUSION: The proposed deep learning segmentation algorithm (CNN) for automated eye compartment segmentation in OCT B-scans (SDOCT and SSOCT) is on par with manual segmentations by human graders.

Chlorine Dioxide-Pollutant Transformation and Formation of Hypochlorous Acid as a Secondary Oxidant.

Chlorine dioxide (ClO2) has been used as a disinfectant in water treatment for a long time, and its use for micropollutant abatement in wastewater has recently been suggested. Surprisingly, a mechanistic understanding of ClO2 reactions in (waste)water matrices is largely lacking. The present study contributes to this mechanistic understanding by performing a detailed investigation of ClO2 reactions with organic matter using phenol as a surrogate for reactive phenolic moieties. A concept for indirectly determining HOCl using 2- and 4-bromophenol was developed. The reaction of phenol with ClO2 formed chlorite (62 ± 4% per ClO2 consumed) and hypochlorous acid (HOCl) (42 ± 3% per ClO2 consumed). The addition of ClO2 to wastewater (5 × 10-5 M ClO2) resulted in 40% atenolol and 47% metoprolol transformation. The presence of the selective HOCl scavenger glycine largely diminished their transformation, indicating that atenolol and metoprolol were transformed by a fast reaction with HOCl (e.g., k (atenolol + HOCl) = 3.5 × 104 M-1 s-1) that formed in ClO2 reactions with the wastewater matrix. The formation of HOCl may thus increase the number of transformable micropollutants in ClO2 applications. However, chlorine related byproducts may also be formed.

Comparison of contact radiographed and stained histological sections for osseointegration analysis of dental implants: an in vivo study.

OBJECTIVES: Histology is still regarded as the gold-standard to determine bone implant contact (BIC) as a parameter representing implant stability. As the further processing of cut slices for contact radiography (CR) to stained and polished histological sections is time consuming and error prone, our aim was to assess agreement between CR and Giemsa-eosin (GE) stained sections with regard to dental implants. STUDY DESIGN: Threaded dental titanium implants (n = 54) from the maxillae of Goettingen minipigs were evaluated. After 28 and 56 days, BIC and the ratio of bone volume to total volume (BV/TV; 1000 µm) were determined on the same sections by using CR and GE staining, and the results were compared. RESULTS: Moderate differences for BIC (0.6%; P = .53) and BV/TV (1.3%; P = .01) between the methods were determined, in which CR overestimated BIC and BV/TV. A strong correlation was seen between the modalities concerning BIC (28 days: r = 0.84; 56 days: r = 0.85; total: r = 0.85) and BV/TV (r = 0.96; r = 0.94; r = 0.96; all: P < .0001). CONCLUSIONS: CR enabled determination of the bone-to-implant interface in comparison with GE-stained sections. BIC and BV/TV were slightly overestimated but correlated strongly between the methods. Therefore, if BIC and BV/TV are sufficient endpoints, CR is adequate and no further preparation and staining are necessary.

Recent loss of self-incompatibility by degradation of the male component in allotetraploid Arabidopsis kamchatica.

The evolutionary transition from outcrossing to self-fertilization (selfing) through the loss of self-incompatibility (SI) is one of the most prevalent events in flowering plants, and its genetic basis has been a major focus in evolutionary biology. In the Brassicaceae, the SI system consists of male and female specificity genes at the S-locus and of genes involved in the female downstream signaling pathway. During recent decades, much attention has been paid in particular to clarifying the genes responsible for the loss of SI. Here, we investigated the pattern of polymorphism and functionality of the female specificity gene, the S-locus receptor kinase (SRK), in allotetraploid Arabidopsis kamchatica. While its parental species, A. lyrata and A. halleri, are reported to be diploid and mainly self-incompatible, A. kamchatica is self-compatible. We identified five highly diverged SRK haplogroups, found their disomic inheritance and, for the first time in a wild allotetraploid species, surveyed the geographic distribution of SRK at the two homeologous S-loci across the species range. We found intact full-length SRK sequences in many accessions. Through interspecific crosses with the self-incompatible and diploid congener A. halleri, we found that the female components of the SI system, including SRK and the female downstream signaling pathway, are still functional in these accessions. Given the tight linkage and very rare recombination of the male and female components on the S-locus, this result suggests that the degradation of male components was responsible for the loss of SI in A. kamchatica. Recent extensive studies in multiple Brassicaceae species demonstrate that the loss of SI is often derived from mutations in the male component in wild populations, in contrast to cultivated populations. This is consistent with theoretical predictions that mutations disabling male specificity are expected to be more strongly selected than mutations disabling female specificity, or the female downstream signaling pathway.