Set Of Images Research Articles

Characterization of Partial Discharges in Dielectric Oils Using High-Resolution CMOS Image Sensor and Convolutional Neural Networks.

In this work, an exhaustive analysis of the partial discharges that originate in the bubbles present in dielectric mineral oils is carried out. To achieve this, a low-cost, high-resolution CMOS image sensor is used. Partial discharge measurements using that image sensor are validated by a standard electrical detection system that uses a discharge capacitor. In order to accurately identify the images corresponding to partial discharges, a convolutional neural network is trained using a large set of images captured by the image sensor. An image classification model is also developed using deep learning with a convolutional network based on a TensorFlow and Keras model. The classification results of the experiments show that the accuracy achieved by our model is around 95% on the validation set and 82% on the test set. As a result of this work, a non-destructive diagnosis method has been developed that is based on the use of an image sensor and the design of a convolutional neural network. This approach allows us to obtain information about the state of mineral oils before breakdown occurs, providing a valuable tool for the evaluation and maintenance of these dielectric oils.

RADCURE: An open-source head and neck cancer CT dataset for clinical radiation therapy insights.

This manuscript presents RADCURE, one of the most extensive head and neck cancer (HNC) imaging datasets accessible to the public. Initially collected for clinical radiation therapy (RT) treatment planning, this dataset has been retrospectively reconstructed for use in imaging research. RADCURE encompasses data from 3346 patients, featuring computed tomography (CT) RT simulation images with corresponding target and organ-at-risk contours. These CT scans were collected using systems from three different manufacturers. Standard clinical imaging protocols were followed, and contours were manually generated and reviewed at weekly RT quality assurance rounds. RADCURE imaging and structure set data was extracted from our institution's radiation treatment planning and oncology information systems using a custom-built data mining and processing system. Furthermore, images were linked to our clinical anthology of outcomes data for each patient and includes demographic, clinical and treatment information based on the 7th edition TNM staging system (Tumor-Node-Metastasis Classification System of Malignant Tumors). The median patient age is 63, with the final dataset including 80% males. Half of the cohort is diagnosed with oropharyngeal cancer, while laryngeal, nasopharyngeal, and hypopharyngeal cancers account for 25%, 12%, and 5% of cases, respectively. The median duration of follow-up is five years, with 60% of the cohort surviving until the last follow-up point. The dataset provides images and contours in DICOM CT and RT-STRUCT formats, respectively. We have standardized the nomenclature for individual contours-such as the gross primary tumor, gross nodal volumes, and 19 organs-at-risk-to enhance the RT-STRUCT files' utility. Accompanying demographic, clinical, and treatment data are supplied in a comma-separated values (CSV) file format. This comprehensive dataset is publicly accessible via The Cancer Imaging Archive. RADCURE's amalgamation of imaging, clinical, demographic, and treatment data renders it an invaluable resource for a broad spectrum of radiomics image analysis research endeavors. Researchers can utilize this dataset to advance routine clinical procedures using machine learning or artificial intelligence, to identify new non-invasive biomarkers, or to forge prognostic models.

Learning-based surface deformation recovery for large radio telescope antennas

Abstract The surface deformation of the main reflector in a large radio telescope is closely related to its working efficiency, which is important for some astronomical science studies. Here, we present a deep learning-based surface deformation recovery framework using non-interferometric intensity measurements as input. The recurrent convolutional neural network (RCNN) is developed to establish the inverse mapping relationship between the surface deformation of the main reflector and the intensity images at the aperture plane and at a near-field plane. Meanwhile, a physical forward propagation model is adopted to generate a large amount of data for pre-training in a computationally efficient manner. Then, the inverse mapping relationship is adjusted and improved by transfer learning using experimental data, which achieves a 15-fold reduction in the number of training image sets required, which is helpful to facilitate the practical application of deep learning in this field. In addition, the RCNN model can be trained as a denoiser, and it is robust to the axial positioning error of the measuring points. It is also promising to extend this method to the study of adaptive optics.

ERGO-ML: comparing IllustrisTNG and HSC galaxy images via contrastive learning

ABSTRACT Modern cosmological hydrodynamical galaxy simulations provide tens of thousands of reasonably realistic synthetic galaxies across cosmic time. However, quantitatively assessing the level of realism of simulated universes in comparison to the real one is difficult. In this paper of the Extracting Reality from Galaxy Observables with Machine Learning series, we utilize contrastive learning to directly compare a large sample of simulated and observed galaxies based on their stellar-light images. This eliminates the need to specify summary statistics and allows to exploit the whole information content of the observations. We produce survey-realistic galaxy mock data sets resembling real Hyper Suprime-Cam (HSC) observations using the cosmological simulations TNG50 and TNG100. Our focus is on galaxies with stellar masses between 109 and 1012 M⊙ at z = 0.1–0.4. This allows us to evaluate the realism of the simulated TNG galaxies in comparison to actual HSC observations. We apply the self-supervised contrastive learning method Nearest Neighbour Contrastive Learning to the images from both simulated and observed data sets (g-, r-, i-bands). This results in a 256-dimensional representation space, encoding all relevant observable galaxy properties. First, this allows us to identify simulated galaxies that closely resemble real ones by seeking similar images in this multidimensional space. Even more powerful, we quantify the alignment between the representations of these two image sets, finding that the majority (≳ 70 per cent) of the TNG galaxies align well with observed HSC images. However, a subset of simulated galaxies with larger sizes, steeper Sérsic profiles, smaller Sérsic ellipticities, and larger asymmetries appears unrealistic. We also demonstrate the utility of our derived image representations by inferring properties of real HSC galaxies using simulated TNG galaxies as the ground truth.

Using Computer Vision to Detect E-cigarette Content in TikTok Videos.

Previous research has identified abundant e-cigarette content on social media using primarily text-based approaches. However, frequently used social media platforms among youth, such as TikTok, contain primarily visual content, requiring the ability to detect e-cigarette-related content across large sets of videos and images. This study aims to use a computer vision technique to detect e-cigarette-related objects in TikTok videos. We searched 13 hashtags related to vaping on TikTok (eg, #vape) in November 2022 and obtained 826 still images extracted from a random selection of 254 posts. We annotated images for the presence of vaping devices, hands, and/or vapor clouds. We developed a YOLOv7-based computer vision model to detect these objects using 85% of extracted images (N = 705) for training and 15% (N = 121) for testing. Our model's recall value was 0.77 for all three classes: vape devices, hands, and vapor. Our model correctly classified vape devices 92.9% of the time, with an average F1 score of 0.81. The findings highlight the importance of having accurate and efficient methods to identify e-cigarette content on popular video-based social media platforms like TikTok. Our findings indicate that automated computer vision methods can successfully detect a range of e-cigarette-related content, including devices and vapor clouds, across images from TikTok posts. These approaches can be used to guide research and regulatory efforts. Object detection, a computer vision machine learning model, can accurately and efficiently identify e-cigarette content on a primarily visual-based social media platform by identifying the presence of vaping devices and evidence of e-cigarette use (eg, hands and vapor clouds). The methods used in this study can inform computational surveillance systems for detecting e-cigarette content on video- and image-based social media platforms to inform and enforce regulations of e-cigarette content on social media.

Predictive cues elicit a liminal confirmation bias in the moral evaluation of real-world images.

Previous research suggested that predictive cues enhance the preference and reduce the response time for congruent targets during bivalent food evaluation, indicating a confirmation bias. Less is known about how prior processing affects subjective moral evaluation. Here, we used three different types of predictive cues to elicit directional vs. non-predictive prior processing and then asked the participants to perform moral evaluations on a continuous scale from -10 ("very immoral") to +10 ("very moral") with a diverse set of real-world images. Our experimental image database balanced the morality of image content and the volatility of the ratings based on the means and standard deviations in a preliminary study. Ratings, response times, and gaze positions were measured to examine the effects of predictive cues on the moral evaluation of real-world images. We found that the moral ratings were in line with the expectations induced by the cues. Compared to the non-predictive condition, the moral evaluation in the directional conditions was more polarized. For neutral images, the predictive cues tilted the evaluations to positive vs. negative, indicating a decisive liminal influence. High-volatility images were impacted more than low-volatility images in ratings as well as response times. Furthermore, the gaze positions during the interval between the predictive cue and the image showed a spatial displacement in line with the cue instruction, indicating a response bias. Together, the results show that predictive cues elicit a liminal confirmation bias in moral image evaluation, much in the same way as in bivalent food evaluation.

Impact of point-of-care ultrasound use on patient referral decisions in rural Kenya: a mixed methods study

BackgroundPoint-of-care ultrasound (POCUS) is recognized as a key imaging modality to bridge the diagnostic imaging gap in Low- and Middle-Income Countries (LMICs). POCUS use has been shown to impact patient management decisions including referral for specialist care. This study explored the impact of POCUS use on referral decisions among trained healthcare providers working in primary rural and peri-urban health facilities in Kenya.MethodsA concurrent mixed methods approach was used, including a locally developed survey (N = 38) and semi-structured interviews of POCUS trained healthcare providers (N = 12). Data from the survey was descriptively analyzed and interviews were evaluated through the framework matrix method.ResultsSurvey results of in-facility access to Xray, Ultrasonography, CT scan and MRI were 49%, 33%, 3% and 0% respectively. Only 54% of the facilities where trainees worked had the capacity to perform cesarean sections, and 38% could perform general surgery. Through a combined inductive and deductive evaluation of interview data, we found that the emerging themes could be organized through the framework of the six domains of healthcare quality as described by the Institute of Medicine: Providers reported that POCUS use allowed them to make referral decisions which were timely, safe, effective, efficient, equitable and patient-centered. Challenges included machine breakdown, poor image quality, practice isolation, lack of institutional support and insufficient feedback on the condition of patients after referral.ConclusionThis study highlighted that in the setting of limited imaging and surgical capacity, POCUS use by trained providers in Kenyan primary health facilities has the potential to improve the patient referral process and to promote key dimensions of healthcare quality. Therefore, there is a need to expand POCUS training programs and to develop context specific POCUS referral algorithms.

Clinical efficacy of motion-insensitive imaging technique with deep learning reconstruction to improve image quality in cervical spine MR imaging.

To demonstrate that a T2 periodically rotated overlapping parallel lines with enhanced reconstruction (PROPELLER) technique using deep learning reconstruction (DLR) will provide better image quality and decrease image noise. From December 2020 to March 2021, 35 patients examined cervical spine MRI were included in this study. Four sets of images including fast spin echo (FSE), original PROPELLER, PROPELLER DLR50%, and DLR75% were quantitatively and qualitatively reviewed. We calculated the signal-to-noise ratio (SNR) of the spinal cord and sternocleidomastoid (SCM) muscle and the contrast-to-noise ratio (CNR) of the spinal cord by applying region-of-interest at the spinal cord, SCM muscle, and background air. We evaluated image noise with regard to the spinal cord, SCM, and back muscles at each level from C2-3 to C6-7 in the 4 sets. At all disc levels, the mean SNR values for the spinal cord and SCM muscles were significantly higher in PROPELLER DLR50% and DLR75% compared to FSE and original PROPELLER images (P < .0083). The mean CNR values of the spinal cord were significantly higher in PROPELLER DLR50% and DLR75% compared to FSE at the C3-4 and 4-5 levels and PROPELLER DLR75% compared to FSE at the C6-7 level (P < .0083). Qualitative analysis of image noise on the spinal cord, SCM, and back muscles showed that PROPELLER DLR50% and PROPELLER DLR75% images showed a significant denoising effect compared to the FSE and original PROPELLER images. The combination of PROPELLER and DLR improved image quality with a high SNR and CNR and reduced noise. Motion-insensitive imaging technique (PROPELLER) increased the image quality compared to conventional FSE images. PROPELLER technique with a DLR reduced image noise and improved image quality.

BEYOND GROUND CONTROL POINTS: COST-EFFECTIVE 3D BUILDING RECONSTRUCTION THROUGH GNSS-INTEGRATED PHOTOGRAMMETRY

Abstract. The process of 3D building modeling serves a multitude of practical and strategic purposes across diverse industries. Building a 3D model involves employing a range of techniques and technologies. Among these, the most used methods include 3D laser scanning and photogrammetry, whether applied at close-range or through the use of Unmanned Aerial Systems (UAS). In photogrammetry, ground control points (GCPs) are generally needed to scale and georeference the digital reconstruction process, but it is a timeconsuming practice or sometimes impractical or dangerous. This paper aims to evaluate the efficiency of two integrated devices to perform photogrammetric 3D reconstruction without GCPs. They are both composed by a Sony ZV1 camera coupled with two different RTK/PPK GNSS system: the Emlid Reach RS2 GNSS receiver and the Emlid Reach M2 module with a multi-band GNSS helical antenna. Different sets of images were acquired with the two proposed devices for the lever-arm estimation and to perform the 3D surveying of the Galata monastery historical monument. The accuracy of the process and derived dense point clouds is assessed by comparing them with GCPs and a reference point cloud derived by fusing an UAS and a high-resolution mobile laser scanning point cloud. The ultimate goal is to obtain a 3D building model without the use of GCPs in the process of bundle block adjustment with centimeter accuracy.

Deep hybrid manifold for image set classification

The exponential growth of the data volume of image sets, which contain more information than a single image, has attracted increasing attention from researchers. Image set data are often described as covariance matrices or linear subspaces, and the unique geometries they span are symmetric positive definite (SPD) manifolds and Grassmann manifolds, respectively. Image set data are often described as covariance matrices or linear subspaces, and the distinctive geometries they span are symmetric positive definite (SPD) manifold and Grassmann manifold, respectively. However, most studies focus on a single manifold and ignore the useful information of the another manifold. Based on this, we propose a new Deep Hybrid Manifold Network (DHMNet).The DHMNet consists of backbone network, stackable Hybrid Manifold AutoEncoder (HMAE) and,Maximum Fusion Module (MFM). The image set data is modeled through SPD manifold and Grassmann manifold. The modeled data is input into the backbone network composed of SPDNet and GrNet for initial feature extraction, and the output manifold data are input into HMAEs. The HMAE effectively extracts and hybridizes complementary information from different manifolds and has the ability to generate deep representations with rich structural semantic information. For the three image datasets used, DHMNet with two HMAEs improves the classification accuracy by 3.83–5.76% over the classical SPDNet, and even reaches the best when compared to other models, with the best performance on the First Person Hand Action (FPHA) dataset for skeleton-based hand action recognition.

Ketogenic diet reduces a neurobiological craving signature in inpatients with alcohol use disorder.

Increasing evidence suggests that a ketogenic (high-fat, low-carbohydrate) diet (KD) intervention reduces alcohol withdrawal severity and alcohol craving in individuals with alcohol use disorder (AUD) by shifting brain energetics from glucose to ketones. We hypothesized that the KD would reduce a neurobiological craving signature when individuals undergoing alcohol detoxification treatment were exposed to alcohol cues. We performed a secondary analysis of functional magnetic resonance data of 33 adults with an AUD who were randomized to a KD (n = 19) or a standard American diet (SA; n = 14) and underwent 3 weeks of inpatient alcohol detoxification treatment. Once per week, participants performed an alcohol cue-reactivity paradigm with functional magnetic resonance imaging. We extracted brain responses to food and alcohol cues and quantified the degree to which each set of brain images shared a pattern of activation with a recently established 'Neurobiological Craving Signature' (NCS). We then performed a group-by-time repeated measures ANOVA to test for differences in craving signature expression between the dietary groups over the three-week treatment period. We also correlated these expression patterns with self-reported wanting ratings for alcohol cues. For alcohol relative to food cues, there was a main effect of group, such that the KD group showed lower NCS expression across all 3 weeks of treatment. The main effect of time and the group-by-time interaction were not significant. Self-reported wanting for alcohol cues reduced with KD compared to SA but did not correlate with the NCS score. A ketogenic diet reduces self-reported alcohol wanting, and induced lower NCS to alcohol cues during inpatient treatment for AUD. However, in the KD group alcohol wanting continued to decrease across the 3 weeks of abstinence while the NCS scores remained stable, suggesting that this cue-induced NCS may not fully capture ongoing, non-cue-induced alcohol desire.

Pooled Steganalysis via Model Discrepancy

Pooled steganalysis aims to discover the guilty actor(s) among multiple normal actor(s). Existing techniques mainly rely on the high-dimension and time-consuming features. Moreover, the minor feature distance between cover and stego is detrimental to pooled steganalysis. To overcome these issues, this paper focuses on the discrepancy of the statistical characteristics of transmitted multiple images and designs a model-based effective pooled steganalysis strategy. Facing the public and monitored channel, without using the feature extractions, pooled steganalysis collects a set of images transmitted by a suspicious actor and use the corresponding distortion values as the statistic representation of the selected image set. Specifically, the normalized distortion of the suspicious image set generated via normal/guilty actor(s) is modelled as a normal distribution, and we apply maximum likelihood estimation (MLE) to estimate the parameter (cluster center) of the distribution by which we can represent the defined model. Considering the tremendous distortion difference between normal and stego image sets, we can deduce that the constructed model can effectively discover and reveal the existence of abnormal behavior of guilty actors. To show the discrepancy of different models, employing the logistic function and likelihood ratio test (LRT), we construct a new detector by which the ratio of cluster centers is turned into a probability. Depending on the generated probability and an optimal threshold, we make a judgment on whether the dubious actor is normal or guilty. Extensive experiments demonstrate that, compared to existing pooled steganalysis techniques, the proposed scheme exhibits great detection performance on the guilty actor(s) with lower complexity.

Deep learning-accelerated T2-weighted imaging versus conventional T2-weighted imaging in the female pelvic cavity: image quality and diagnostic performance.

The deep learning (DL)-based reconstruction algorithm reduces noise in magnetic resonance imaging (MRI), thereby enabling faster MRI acquisition. To compare the image quality and diagnostic performance of conventional turbo spin-echo (TSE) T2-weighted (T2W) imaging with DL-accelerated sagittal T2W imaging in the female pelvic cavity. This study evaluated 149 consecutive female pelvic MRI examinations, including conventional T2W imaging with TSE (acquisition time = 2:59) and DL-accelerated T2W imaging with breath hold (DL-BH) (1:05 [0:14 × 3 breath-holds]) in the sagittal plane. In 294 randomly ordered sagittal T2W images, two radiologists independently assessed image quality (sharpness, subjective noise, artifacts, and overall image quality), made a diagnosis for uterine leiomyomas, and scored diagnostic confidence. For the uterus and piriformis muscle, quantitative imaging analysis was also performed. Wilcoxon signed rank tests were used to compare the two sets of T2W images. In the qualitative analysis, DL-BH showed similar or significantly higher scores for all features than conventional T2W imaging (P <0.05). In the quantitative analysis, the noise in the uterus was lower in DL-BH, but the noise in the muscle was lower in conventional T2W imaging. In the uterus and muscle, the signal-to-noise ratio was significantly lower in DL-BH than in conventional T2W imaging (P <0.001). The diagnostic performance of the two sets of T2W images was not different for uterine leiomyoma. DL-accelerated sagittal T2W imaging obtained with three breath-holds demonstrated superior or comparable image quality to conventional T2W imaging with no significant difference in diagnostic performance for uterine leiomyomas.

Using Remote Sensing Multispectral Imagery for Invasive Species Quantification: The Effect of Image Resolution on Area and Biomass Estimation

This study assesses the applicability of different-resolution multispectral remote sensing images for mapping and estimating the aboveground biomass (AGB) of Carpobrotus edulis, a prominent invasive species in European coastal areas. This study was carried out on the Cávado estuary sand spit (Portugal). The performance of three sets of multispectral images with different Ground Sample Distances (GSDs) were compared: 2.5 cm, 5 cm, and 10 cm. The images were classified using the supervised classification algorithm random forest and later improved by applying a sieve filter. Samples of C. edulis were also collected, dried, and weighed to estimate the AGB using the relationship between the dry weight (DW) and vegetation indices (VIs). The resulting regression models were evaluated based on their coefficient of determination (R2), Normalised Root Mean Square Error (NRMSE), p-value, Akaike information criterion (AIC), and the Bayesian information criterion (BIC). The results show that the three tested image resolutions allow for constructing reliable coverage maps of C. edulis, with overall accuracy values of 89%, 85%, and 88% for the classification of the 2.5 cm, 5 cm, and 10 cm GSD images, respectively. The best-performing VI-DW regression models achieved R2 = 0.87 and NRMSE = 0.09 for the 2.5 cm resolution; R2 = 0.77 and NRMSE = 0.12 for the 5 cm resolution; and R2 = 0.64 and NRMSE = 0.15 for the 10 cm resolution. The C. edulis area and total AGB were 3441.10 m2 and 28,327.1 kg (with an AGB relative error (RE) = 0.08) for the 2.5 cm resolution; 3070.04 m2 and 29,170.8 kg (AGB RE = 0.08) for the 5 cm resolution; and 2305.06 m2 and 22,135.7 kg (AGB RE = 0.11) for the 10 cm resolution. Spatial and model differences were analysed in detail to determine their causes. Final analyses suggest that multispectral imagery of up to 5 cm GSD is adequate for estimating C. edulis distribution and biomass.

Diagnosis of malignancy in oropharyngeal confocal laser endomicroscopy using GPT 4.0 with vision.

Confocal Laser Endomicroscopy (CLE) is an imaging tool, that has demonstrated potential for intraoperative, real-time, non-invasive, microscopical assessment of surgical margins of oropharyngeal squamous cell carcinoma (OPSCC). However, interpreting CLE images remains challenging. This study investigates the application of OpenAI's Generative Pretrained Transformer (GPT) 4.0 with Vision capabilities for automated classification of CLE images in OPSCC. CLE Images of histological confirmed SCC or healthy mucosa from a database of 12 809 CLE images from 5 patients with OPSCC were retrieved and anonymized. Using a training data set of 16 images, a validation set of 139 images, comprising SCC (83 images, 59.7%) and healthy normal mucosa (56 images, 40.3%) was classified using the application programming interface (API) of GPT4.0. The same set of images was also classified by CLE experts (two surgeons and one pathologist), who were blinded to the histology. Diagnostic metrics, the reliability of GPT and inter-rater reliability were assessed. Overall accuracy of the GPT model was 71.2%, the intra-rater agreement was κ = 0.837, indicating an almost perfect agreement across the three runs of GPT-generated results. Human experts achieved an accuracy of 88.5% with a substantial level of agreement (κ = 0.773). Though limited to a specific clinical framework, patient and image set, this study sheds light on some previously unexplored diagnostic capabilities of large language models using few-shot prompting. It suggests the model`s ability to extrapolate information and classify CLE images with minimal example data. Whether future versions of the model can achieve clinically relevant diagnostic accuracy, especially in uncurated data sets, remains to be investigated.

Optimizing skin disease diagnosis: harnessing online community data with contrastive learning and clustering techniques

Skin diseases pose significant challenges in China. Internet health forums offer a platform for millions of users to discuss skin diseases and share images for early intervention, leaving large amount of valuable dermatology images. However, data quality and annotation challenges limit the potential of these resources for developing diagnostic models. In this study, we proposed a deep-learning model that utilized unannotated dermatology images from diverse online sources. We adopted a contrastive learning approach to learn general representations from unlabeled images and fine-tuned the model on coarsely annotated images from Internet forums. Our model classified 22 common skin diseases. To improve annotation quality, we used a clustering method with a small set of standardized validation images. We tested the model on images collected by 33 experienced dermatologists from 15 tertiary hospitals and achieved a 45.05% top-1 accuracy, outperforming the published baseline model by 3%. Accuracy increased with additional validation images, reaching 49.64% with 50 images per category. Our model also demonstrated transferability to new tasks, such as detecting monkeypox, with a 61.76% top-1 accuracy using only 50 additional images in the training process. We also tested our model on benchmark datasets to show the generalization ability. Our findings highlight the potential of unannotated images from online forums for future dermatology applications and demonstrate the effectiveness of our model for early diagnosis and potential outbreak mitigation.

Multimodal Biomedical Image Segmentation using Multi-Dimensional U-Convolutional Neural Network

Deep learning recently achieved advancement in the segmentation of medical images. In this regard, U-Net is the most predominant deep neural network, and its architecture is the most prevalent in the medical imaging society. Experiments conducted on difficult datasets directed us to the conclusion that the traditional U-Net framework appears to be deficient in certain respects, despite its overall excellence in segmenting multimodal medical images. Therefore, we propose several modifications to the existing cutting-edge U-Net model. The technical approach involves applying a Multi-Dimensional U-Convolutional Neural Network to achieve accurate segmentation of multimodal biomedical images, enhancing precision and comprehensiveness in identifying and analyzing structures across diverse imaging modalities. As a result of the enhancements, we propose a novel framework called Multi-Dimensional U-Convolutional Neural Network (MDU-CNN) as a potential successor to the U-Net framework. On a large set of multimodal medical images, we compared our proposed framework, MDU-CNN, to the classical U-Net. There have been small changes in the case of perfect images, and a huge improvement is obtained in the case of difficult images. We tested our model on five distinct datasets, each of which presented unique challenges, and found that it has obtained a better performance of 1.32%, 5.19%, 4.50%, 10.23% and 0.87%, respectively.

Thickness and surface estimation of thin porous media based on 3D image data

Measuring the thickness of thin porous materials provides valuable insights into their structure, properties, and performance, including key properties such as porosity and permeability, and is highly beneficial for a range of industrial applications, particularly for ensuring effective quality control processes. A novel approach for estimating the thickness of porous media and their surfaces is proposed based on voxel sets of 3D images, such as 3D scans and segmented scan data. Initially, the solid volume fraction (SVF) is computed for each voxel layer perpendicular to the through direction. Then, fitting functions consisting of piecewise linear segments are chosen to ensure an accurate representation of the layer data. Each function is associated with various thickness regions of the medium, including the medium itself and its surface. An optimization problem is then solved to find the best-fitting function based on the squared area between the SVF and the fitting function. The thickness of the medium and its surfaces is determined based on the identified optimal fit. This robust, reliable, and fast approach aims to provide not only a non-intrusive method for thickness estimation of porous media represented by voxel sets but also a precise alternative to existing methodologies.

Supporting Safety Analysis of Image-processing DNNs through Clustering-based Approaches

The adoption of deep neural networks (DNNs) in safety-critical contexts is often prevented by the lack of effective means to explain their results, especially when they are erroneous. In our previous work, we proposed a white-box approach (HUDD) and a black-box approach (SAFE) to automatically characterize DNN failures. They both identify clusters of similar images from a potentially large set of images leading to DNN failures. However, the analysis pipelines for HUDD and SAFE were instantiated in specific ways according to common practices, deferring the analysis of other pipelines to future work. In this paper, we report on an empirical evaluation of 99 different pipelines for root cause analysis of DNN failures. They combine transfer learning, autoencoders, heatmaps of neuron relevance, dimensionality reduction techniques, and different clustering algorithms. Our results show that the best pipeline combines transfer learning, DBSCAN, and UMAP. It leads to clusters almost exclusively capturing images of the same failure scenario, thus facilitating root cause analysis. Further, it generates distinct clusters for each root cause of failure, thus enabling engineers to detect all the unsafe scenarios. Interestingly, these results hold even for failure scenarios that are only observed in a small percentage of the failing images.

A new bio imagery user-friendly tool for automatic morphometry measurement on muscle cell cultures and histological sections

TRUEFAD (TRUE Fiber Atrophy Distinction) is a bioimagery user-friendly tool developed to allow consistent and automatic measurement of myotube diameter in vitro, muscle fiber size and type using rodents and human muscle biopsies. This TRUEFAD package was set up to standardize and dynamize muscle research via easy-to-obtain images run on an open-source plugin for FIJI. We showed here both the robustness and the performance of our pipelines to correctly segment muscle cells and fibers. We evaluated our pipeline on real experiment image sets and showed consistent reliability across images and conditions. TRUEFAD development makes possible systematical and rapid screening of substances impacting muscle morphology for helping scientists focus on their hypothesis rather than image analysis.