Automatic Characterization Research Articles

An end-to-end framework for intima media measurement and atherosclerotic plaque detection in the carotid artery

Background and objectives: The detection and delineation of atherosclerotic plaque are usually manually performed by medical experts on the carotid artery. Evidence suggests that this manual process is subject to errors and has a large variability between experts, equipment, and datasets. This paper proposes a robust end-to-end framework for automatic atherosclerotic plaque detection.Methods: The proposed framework is composed of: (1) a semantic segmentation model based on U-Net, with EfficientNet as the backbone, that obtains a segmentation mask with the carotid intima-media region; and (2) a convolutional neural network designed using Bayesian optimization that simultaneously performs a regression to get the average and maximum carotid intima media thickness, and a classification to determine the presence of plaque.Results: Our approach improves the state-of-the-art in both co and bulb territories in the REGICOR database, with more than 8000 images, while providing predictions in real-time. The correlation coefficient was 0.89 in the common carotid artery and 0.74 for bulb region, and the F1 score for atherosclerotic plaque detecting was 0.60 and 0.59, respectively. The experimentation carried out includes a comparison with other fully automatic methods for carotid intima media thickness estimation found in the literature. Additionally, we present an extensive experimental study to evaluate the robustness of our proposal, as well as its suitability and efficiency compared to different versions of the framework.Conclusions: The proposed end-to-end framework significantly improves the automatic characterization of atherosclerotic plaque. The generation of the segmented mask can be helpful for practitioners since it allows them to evaluate and interpret the model’s results by visual inspection. Furthermore, the proposed framework overcomes the limitations of previous research based on ad-hoc post-processing, which could lead to overestimations in the case of oblique forms of the carotid artery.

Discovering Key Topics in Emergency Medical Dispatch from Free Text Dispatcher Observations.

The objective of this work was to discover key topics latent in free text dispatcher observations registered during emergency medical calls. We used a total of 1374931 independent retrospective cases from the Valencian emergency medical dispatch service in Spain, from 2014 to 2019. Text fields were preprocessed to reduce vocabulary size and filter noise, removing accent and punctuation marks, along with uninformative and infrequent words. Key topics were inferred from the multinomial probabilities over words conditioned on each topic from a Latent Dirichlet Allocation model, trained following an online mini-batch variational approach. The optimal number of topics was set analyzing the values of a topic coherence measure, based on the normalized pointwise mutual information, across multiple validation K-folds. Our results support the presence of 15 key topics latent in free text dispatcher observations, related with: ambulance request; chest pain and heart attack; respiratory distress; head falls and blows; fever, chills, vomiting and diarrhea; heart failure; syncope; limb injuries; public service body request; thoracic and abdominal pain; stroke and blood pressure abnormalities; pill intake; diabetes; bleeding; consciousness. The discovery of these topics implies the automatic characterization of a huge volume of complex unstructured data containing relevant information linked to emergency medical call incidents. Hence, results from this work could lead to the update of structured emergency triage algorithms to directly include this latent information in the triage process, resulting in a positive impact in patient wellbeing and health services sustainability.

Automatic characterization of human embryos at day 4 post-insemination from time-lapse imaging using supervised contrastive learning and inductive transfer learning techniques

Background: Embryo morphology is a predictive marker for implantation success and ultimately live births. Viability evaluation and quality grading are commonly used to select the embryo with the highest implantation potential. However, the traditional method of manual embryo assessment is time-consuming and highly susceptible to inter- and intra-observer variability. Automation of this process results in more objective and accurate predictions.Method: In this paper, we propose a novel methodology based on deep learning to automatically evaluate the morphological appearance of human embryos from time-lapse imaging. A supervised contrastive learning framework is implemented to predict embryo viability at day 4 and day 5, and an inductive transfer approach is applied to classify embryo quality at both times.Results: Results showed that both methods outperformed conventional approaches and improved state-of-the-art embryology results for an independent test set. The viability result achieved an accuracy of 0.8103 and 0.9330 and the quality results reached values of 0.7500 and 0.8001 for day 4 and day 5, respectively. Furthermore, qualitative results kept consistency with the clinical interpretation.Conclusions: The proposed methods are up to date with the artificial intelligence literature and have been proven to be promising. Furthermore, our findings represent a breakthrough in the field of embryology in that they study the possibilities of embryo selection at day 4. Moreover, the grad-CAMs findings are directly in line with embryologists’ decisions. Finally, our results demonstrated excellent potential for the inclusion of the models in clinical practice.

Automated Segmentation of Olivine Phenocrysts in a Volcanic Rock Thin Section Using a Fully Convolutional Neural Network

An example of automated characterization and interpretation of the textural and compositional characteristics of solids phases in thin sections using machine learning (ML) is presented. In our study, we focus on the characterization of olivine in volcanic rocks, which is a phase that is often chemically zoned with variable Mg/(Mg + Fe) ratios, so-called magnesian number or mg#. As the olivine crystals represent only less than 10 vol% of the volcanic rock, a pre-processing step is necessary to automatically detect the phases of interest in the images on a pixel level, which is achieved using Deep Learning. A major contribution of the presented approach is to use backscattered electron (BSE) images to: 1) automatically segment all olivine crystals present in the thin section; 2) determine quantitatively their mg#; and 3) identify different populations depending on zoning type (e.g., normal vs reversal zoning) and textural characteristics (e.g., microlites vs phenocrysts). The segmentation of the olivine crystals is implemented with a pretrained fully convolutional neural network model with DeepLabV3 architecture. The model is trained to identify olivine crystals in backscattered electron images using automatically generated training data. The training data are generated automatically from images which can easily be created from X-Ray element maps. Once the olivines are identified in the BSE images, the relationship between BSE intensity value and mg# is determined using a simple regression based on a set of microprobe measurements. This learned functional relationship can then be applied to all olivine pixels of the thin section. If the highest possible map resolution (1 micron per 1 pixel) is selected for the data acquisition, the full processing time of an entire thin section of ∼3×4cm containing more than 1,500 phenocrysts and 20.000 microliths required 140 h of data acquisition (BSE + X-Ray element maps), 8 h of training and 16 h of segmentation and classification. Our further tests demonstrated that the 140 h of data acquisition can be reduced at least by a factor of 4 since only a part of the thin section area (25% or even less) needs to be used for training. The characterization of each additional thin section would only require the BSE data acquisition time (less than 48 h for a whole thin section), without an additional training step. The paper describes the training and processing in detail, shows analytical results and outlines the potential of this Deep Learning approach for petrological applications, resulting in the automatic characterization and interpretation of mineral textures and compositions with an unprecedented high resolution.

Clustering at the Disposal of Industry 4.0: Automatic Extraction of Plant Behaviors.

For two centuries, the industrial sector has never stopped evolving. Since the dawn of the Fourth Industrial Revolution, commonly known as Industry 4.0, deep and accurate understandings of systems have become essential for real-time monitoring, prediction, and maintenance. In this paper, we propose a machine learning and data-driven methodology, based on data mining and clustering, for automatic identification and characterization of the different ways unknown systems can behave. It relies on the statistical property that a regular demeanor should be represented by many data with very close features; therefore, the most compact groups should be the regular behaviors. Based on the clusters, on the quantification of their intrinsic properties (size, span, density, neighborhood) and on the dynamic comparisons among each other, this methodology gave us some insight into the system’s demeanor, which can be valuable for the next steps of modeling and prediction stages. Applied to real Industry 4.0 data, this approach allowed us to extract some typical, real behaviors of the plant, while assuming no previous knowledge about the data. This methodology seems very promising, even though it is still in its infancy and that additional works will further develop it.

Employing Deep Learning Neural Networks for Characterizing Dual-Porosity Reservoirs Based on Pressure Transient Tests

Abstract The deep learning model constituting two neural network models (i.e., densely connected and long short-term memory) has been applied for automatic characterization of dual-porosity reservoirs with infinite, constant pressure, and no-flow external boundaries. A total of 16 different prediction paradigms have been constructed (one classifier to identify the reservoir models and 15 regressors for predicting the dual-porosity reservoir characteristics). Indeed, wellbore storage coefficient, CDe2S, skin factor, interporosity flow coefficient, and storativity ratio have been estimated. The training pressure signals have been simulated using the analytical solution of the governing equations with varying noise percentages. The pressure drop and derivation of the noisy synthetic signals serve as the input signals to the intelligent scenario. The hyperparameters of the intelligent model have been carefully adjusted to improve its prediction performance. The trained classification model attained 99.48% and 99.32% accuracy over the training and testing datasets. The separately trained 15 regressors converged well to estimate the reservoir parameters. The model performance has been demonstrated with three uniquely simulated and real-field cases. The results indicate that the compiled prediction model can accurately identify the reservoir model and estimate the corresponding characteristics.

Automatic characterization of breast lesions using multi-scale attention-guided deep learning of digital histology images

ABSTRACT A multi-scale attention-guided deep learning model is proposed to characterise breast tissue in digital histology images (H&E stained) according to four different histological types including normal, benign, in situ carcinoma and invasive carcinoma. The framework includes two parallel convolutional neural networks with modified VGG16 architecture. The first network analyzes the whole-sample images at low magnification. The second network focuses on the patches extracted from the whole-sample images at high magnification. In the low-magnification network, a global average pooling layer was added at the end of the network to extract the class activation maps for the attention model. A long short-term memory network was adapted as a recurrent attention mechanism to increase the contribution of the relevant parts of each image for classification. In the high-magnification network, the probability vectors were averaged over all patches extracted from an image to obtain the probability vectors associated with the four histological types for each sample. The probability vectors for each sample from the high-magnification network and the attention model were fused using a multilayer perceptron network to generate a classification label. Obtained results on an independent test set demonstrated an average accuracy of 97.5% ± 1.0% for the proposed model. An average accuracy of 94.5%, 93.5%, and 96.3% was obtained, respectively, for the separate high- and low-magnification networks, and the multi-scale model without an attention mechanism. The results suggested that a multi-scale strategy coupled with an attention mechanism can improve the accuracy of deep learning models in classifying digital histology images.

Deep learning tools and modeling to estimate the temporal expression of cell cycle proteins from 2D still images.

Automatic characterization of fluorescent labeling in intact mammalian tissues remains a challenge due to the lack of quantifying techniques capable of segregating densely packed nuclei and intricate tissue patterns. Here, we describe a powerful deep learning-based approach that couples remarkably precise nuclear segmentation with quantitation of fluorescent labeling intensity within segmented nuclei, and then apply it to the analysis of cell cycle dependent protein concentration in mouse tissues using 2D fluorescent still images. First, several existing deep learning-based methods were evaluated to accurately segment nuclei using different imaging modalities with a small training dataset. Next, we developed a deep learning-based approach to identify and measure fluorescent labels within segmented nuclei, and created an ImageJ plugin to allow for efficient manual correction of nuclear segmentation and label identification. Lastly, using fluorescence intensity as a readout for protein concentration, a three-step global estimation method was applied to the characterization of the cell cycle dependent expression of E2F proteins in the developing mouse intestine.

Recognition of the Mental Workloads of Pilots in the Cockpit Using EEG Signals

The commercial flightdeck is a naturally multi-tasking work environment, one in which interruptions are frequent come in various forms, contributing in many cases to aviation incident reports. Automatic characterization of pilots’ workloads is essential to preventing these kind of incidents. In addition, minimizing the physiological sensor network as much as possible remains both a challenge and a requirement. Electroencephalogram (EEG) signals have shown high correlations with specific cognitive and mental states, such as workload. However, there is not enough evidence in the literature to validate how well models generalize in cases of new subjects performing tasks with workloads similar to the ones included during the model’s training. In this paper, we propose a convolutional neural network to classify EEG features across different mental workloads in a continuous performance task test that partly measures working memory and working memory capacity. Our model is valid at the general population level and it is able to transfer task learning to pilot mental workload recognition in a simulated operational environment.

Automated Structural Analysis and Quantitative Characterization of Scar Tissue Using Machine Learning

An analysis of scar tissue is necessary to understand the pathological tissue conditions during or after the wound healing process. Hematoxylin and eosin (HE) staining has conventionally been applied to understand the morphology of scar tissue. However, the scar lesions cannot be analyzed from a whole slide image. The current study aimed to develop a method for the rapid and automatic characterization of scar lesions in HE-stained scar tissues using a supervised and unsupervised learning algorithm. The supervised learning used a Mask region-based convolutional neural network (RCNN) to train a pattern from a data representation using MMDetection tools. The K-means algorithm characterized the HE-stained tissue and extracted the main features, such as the collagen density and directional variance of the collagen. The Mask RCNN model effectively predicted scar images using various backbone networks (e.g., ResNet50, ResNet101, ResNeSt50, and ResNeSt101) with high accuracy. The K-means clustering method successfully characterized the HE-stained tissue by separating the main features in terms of the collagen fiber and dermal mature components, namely, the glands, hair follicles, and nuclei. A quantitative analysis of the scar tissue in terms of the collagen density and directional variance of the collagen confirmed 50% differences between the normal and scar tissues. The proposed methods were utilized to characterize the pathological features of scar tissue for an objective histological analysis. The trained model is time-efficient when used for detection in place of a manual analysis. Machine learning-assisted analysis is expected to aid in understanding scar conditions, and to help establish an optimal treatment plan.

An Approach for the Automatic Characterization of Underwater Dunes in Fluviomarine Context

The identification of underwater landforms represents an important role in the study of the seafloor morphology. In this context, the segmentation and characterization of underwater dunes allow a better understanding of the dynamism of the seafloor, since the formation of these structures is directly related to environmental conditions, such as current, tide, grain size, etc. In addition, it helps to ensure safe navigation, especially in the context of navigation channels requiring periodic maintenance. This paper proposes a novel method to automatically characterize the underwater dunes. Its originality relies on the extraction of morphological descriptors not only related to the dune itself, but also to the fields where the dunes are located. Furthermore, the proposed approach involves the entire surface of the dunes, rather than profiles or group of pixels as generally found in previous works. Considering the surface modelled by a digital bathymetric model (DBM), the salient features of the dunes (i.e., crest line, stoss trough, and lee trough) are first identified using a geomorphometric analysis of the DBM. The individual dunes are built by matching the crest lines with their respective troughs according to an object-oriented approach. Then, a series of morphological descriptors, selected through a literature review, are computed by taking advantage of the dune salient features, surface representation, and spatial distribution in the fields where they are located. The validation of the proposed method has been conducted using more than 1200 dunes in the fluvio-marine context of the Northern Traverse of the Saint Lawrence River.

Recognition of Leaf Disease Using Hybrid Convolutional Neural Network by Applying Feature Reduction.

Agriculture is crucial to the economic prosperity and development of India. Plant diseases can have a devastating influence towards food safety and a considerable loss in the production of agricultural products. Disease identification on the plant is essential for long-term agriculture sustainability. Manually monitoring plant diseases is difficult due to time limitations and the diversity of diseases. In the realm of agricultural inputs, automatic characterization of plant diseases is widely required. Based on performance out of all image-processing methods, is better suited for solving this task. This work investigates plant diseases in grapevines. Leaf blight, Black rot, stable, and Black measles are the four types of diseases found in grape plants. Several earlier research proposals using machine learning algorithms were created to detect one or two diseases in grape plant leaves; no one offers a complete detection of all four diseases. The photos are taken from the plant village dataset in order to use transfer learning to retrain the EfficientNet B7 deep architecture. Following the transfer learning, the collected features are down-sampled using a Logistic Regression technique. Finally, the most discriminant traits are identified with the highest constant accuracy of 98.7% using state-of-the-art classifiers after 92 epochs. Based on the simulation findings, an appropriate classifier for this application is also suggested. The proposed technique’s effectiveness is confirmed by a fair comparison to existing procedures.

Towards an automatic characterization of riverscape development by deep learning

AbstractRiverscapes are under pressure from anthropogenic development, and this challenges the conservation of biodiversity, hydromorphology and land types. To assess changes and restoration potential, an understanding of alteration to rivers overtime is necessary. This can be challenging due to lack of data, shortcomings in methods and data formats that are not easily incorporated into the assessment process. Historical aerial imagery exists for rivers prior to modification, but the manual classification is time‐consuming. Deep learning is increasingly used in image processing, and here we outline how a convolutional neural network can be used to automatically classify black and white aerial imagery from the database of the Norwegian mapping authority into habitat types. It is demonstrated how historical imagery can be used to develop maps that can be processed further in a GIS to evaluate natural versus anthropogenic changes over time.

Automatic Image Characterization of Psoriasis Lesions

Psoriasis is a chronic skin disease that affects 125 million people worldwide and, particularly, 2% of the Spanish population, characterized by the appearance of skin lesions due to a growth of the epidermis that is seven times larger than usual. Its diagnosis and monitoring are based on the use of methodologies for measuring the severity and extent of these spots, and this includes a large subjective component. For this reason, this paper presents an automatic method for characterizing psoriasis images that is divided into four parts: image preparation or pre-processing, feature extraction, classification of the lesions, and the obtaining of parameters. The methodology proposed in this work covers different digital-image processing techniques, namely, marker-based image delimitation, hair removal, nipple detection, lesion contour detection, areal-measurement-based lesion classification, as well as lesion characterization by means of red and white intensity. The results obtained were also endorsed by a professional dermatologist. This methodology provides professionals with a common software tool for monitoring the different existing typologies, which proved satisfactory in the cases analyzed for a set of 20 images corresponding to different types of lesions.

Breast cancer detection in ultrasound imaging

Breast cancer has become one of the most cancers among women in worldwide countries, as well as a leading cause of death. The use of ultrasound images in medical diagnosis and treatment of patients is critical. The success of cancer treatment and outcome is largely dependent on early detection. Imaging modalities such as ultrasonography are used to identify cancer. Ultrasound imaging is noninvasive, widely available, simple to use, and less expensive than other imaging technologies. As a result, ultrasonography is becoming more used as a cancer detection tool. Ultrasound imaging, on the other hand, is prone to noise and speckle artifacts. First, the ultrasound machine's raw picture data extraction is disabled. As a result, the process of recognizing malignant spots is prioritized. Artificial neural networks and other tissue characterization approaches are used. This technique was chosen because categorization and detection systems have greatly increased in their ability to assist medical experts in diagnosis. Manually classifying ultrasound images not only takes a long time and effort. As a result, a neural network classification-based automatic tissue characterization technique is proposed. Finally, the newly developed algorithms can aid specialists in recognizing suspicious aberrant tissue locations.

Classification of lung nodule malignancy in computed tomography imaging utilising generative adversarial networks and semi-supervised transfer learning

The pulmonary nodules' malignancy rating is commonly confined in patient follow-up; examining the nodule's activity is estimated with the Positron Emission Tomography (PET) system or biopsy. However, these strategies are usually after the initial detection of the malignant nodules acquired from the Computed Tomography (CT) scan. In this study, a Deep Learning methodology to address the challenge of the automatic characterisation of Solitary Pulmonary Nodules (SPN) detected in CT scans is proposed.The research methodology is based on Convolutional Neural Networks, which have proven to be excellent automatic feature extractors for medical images. The publicly available CT dataset, called Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI), and a small CT scan dataset derived from a PET/CT system, is considered the classification target. New, realistic nodule representations are generated employing Deep Convolutional Generative Adversarial Networks to circumvent the shortage of large-scale data to train robust CNNs. Besides, a hierarchical CNN called Feature Fusion VGG19 (FF-VGG19) was developed to enhance feature extraction of the CNN proposed by the Visual Geometry Group (VGG). Moreover, the generated nodule images are separated into two classes by utilising a semi-supervised approach, called self-training, to tackle weak labelling due to DC-GAN inefficiencies.The DC-GAN can generate realistic SPNs, as the experts could only distinguish 23 % of the synthetic nodule images. As a result, the classification accuracy of FF-VGG19 on the LIDC-IDRI dataset increases by +7%, reaching 92.07 %, while the classification accuracy on the CT dataset is increased by 5 %, reaching 84,3 %.

Training and evaluating machine learning algorithms for ocean microplastics classification through vibrational spectroscopy

Microplastics are contaminants of emerging concern - not only environmental, but also to human health. Characterizing them is of fundamental importance to evaluate their potential impacts and target specific actions aiming to reduce potential harming effects. This study extends the exploration of machine learning classification algorithms applied to FTIR spectra of microplastics collected at sea. A comparison of successful classification models was made in order to evaluate prediction performance for 13 classes of polymers. A rigorous methodology was applied using a pipeline scheme to avoid bias in the training and selection phases. The application of an oversampling technique also contributed by compensating unbalanceness in the dataset. The log-loss was used as the minimization function target and to assess performance. In our analysis, Support Vector Machine Classifier provides a good relationship between simplicity and performance, for a fast and useful automatic characterization of microplastics.

Adaptation of a microCT facility for automatic recognition of bioceramic‐based sealers in molar root canals aimed at endodontic treatment quality control

Computed tomography (CT) technology has seen tremendous growth since its first medical applications in the early 1970s. Nowadays, high‐spatial‐resolution digital detectors enable the micro‐computed tomography (μCT) technique, which has been introduced to a wide variety of applications, including biomedical sciences like odontological applications. Endodontic treatments stand as one of the most common dental procedures, and the quality of the treatment is essential to prevent and control any potential infection of root canals. This treatment is typically achieved by careful cleaning, shaping, and sealing steps. Often, endodontic treatments fail due to a poor seal, even if a meticulous root canal preparation is carried out. The present work reports on the spectrometric adaptation of a μCT facility for accurate quantification of root canal sealing with bioceramics, aimed at assessing the sealing performance of the new Bio‐C Sealer product. A dedicated irradiation setup along with digital image processing techniques have been developed for the accurate and automatic recognition of dental samples root canal anatomy and bioceramic‐based sealers. The performance of the proposed approaches has proved to be qualitatively and quantitatively robust for automatic characterization of the endodontic treatment sealing quality, satisfactorily identifying and segmenting canal root, gutta‐percha cone, and bioceramics.

Identification of potential rockfall sources using UAV-derived point cloud

Recent advances in remote sensing techniques and computer algorithms allow accurate, abundant, and high-resolution geometric information retrieval for rock mass characterization from 3D point clouds. The automatic application of the extracted information for local scale rockfall susceptibility assessment, where discontinuities characteristics play a major role in rocky slope stability, requires step by step logical procedures. This paper presents a novel methodology to use the extracted discontinuity set characteristics for a local scale rockfall susceptibility assessment, tailored for Uncrewed Aerial Vehicle (UAV) data acquisition. The method consists of 4 steps: (i) 3D slope model reconstruction using UAV digital photogrammetry, (ii) automatic characterization of discontinuity sets, (iii) slope stability analysis, and (iv) susceptibility assessment using a new Rockfall Susceptibility Index. The proposed method was applied to a road cut rocky slope in a mountainous area of the Samaria National Park, in Crete Island, Greece. Visual validation indicates that the areas of higher and moderate rockfall susceptibility on the 3D model of the rocky slope are adjacent to rockfall source areas marked by the presence of fallen blocks on the foot of the slope. The proposed methodological workflow presents novelties related to the use of point clouds for the estimation of the Rock Quality Designation (RQD) index, the visualization of discontinuity set spacing, the evaluation of the persistence and the Slope Mass Rating (SMR) index, as well as the incorporation of the persistence of overhangs into the rockfall susceptibility assessment and visualization.

Cognitive and autonomous QoT-driven optical line controller

In the direction of disaggregated and cognitive optical networks, this work proposes and experimentally tests a vendor-agnostic optical line controller architecture capable of autonomously setting the working point of optical amplifiers to maximize the capacity of a ROADM-to-ROADM (reconfigurable optical add–drop multiplexer) link. From a procedural point of view, once the equipment is installed, the presented software framework performs an automatic characterization of the line, span by span, to abstract the properties of the physical layer. This process requires the exploitation of monitoring devices such as optical channel monitors and optical time domain reflectometers, available, in a future perspective, in each amplification site. On the basis of this information, an optimization algorithm determines the working point of each amplifier to maximize the quality of transmission (QoT) over the entire band. The optical line controller has been experimentally tested in the laboratory using two different control strategies, achieving in both cases a homogeneous QoT for each channel close to the maximum average and an excellent match with respect to emulation results. In this framework, the Gaussian noise simulation in Python (GNPy) open source Python library is used as the physical model for optical propagation through the fiber, and the covariance matrix adaptation evolution strategy is used as an optimization algorithm to identify properties of each fiber span and to maximize the link capacity.