Convolutional Neural Network Framework Research Articles

Research on three-dimensional counting algorithm of high-precision suspended cells based on deep learning

The research on the three-dimensional counting algorithm of suspension cells is very important for the online counting of cells. Suspension cells are usually observed and counted by taking out cell samples under the microscope. This method cannot realize the online counting of cells. At the same time, there will be weak movement of suspended cells in three-dimensional space, making space counting more difficult. Aiming at the above difficulties, a three-dimensional counting algorithm of suspended cells based on deep learning is proposed. The Faster regions with convolutional neural network (R-CNN) framework is used to identify the target cells, the feature pyramid structure is introduced to reduce the feature loss in the fusion process, and the soft non-maximum suppression algorithm is used to remove the bounding box to improve the detection effect. To reduce the manual labeling workload, an automatic labeling method is proposed. The improved Faster RCNN framework achieves the F1-score of more than 98% after identifying cells, and our method is more accurate than other advanced methods. In terms of three-dimensional counting algorithms, the cells are spatially tracked. Cell matching and correction are carried out by extracting multiple feature information of cells, and adhesion cells are extracted separately for priority processing. The algorithm carried out a tracking experiment on the image sequences of five groups of suspended cells, and the final tracking accuracy of the experiment reached more than 95%. The tracking accuracy is better than other advanced algorithms. Experimental results show that the algorithm achieves high-precision three-dimensional tracking and counting of suspended cells.

An intelligent prediagnosis system for disease prediction and examination recommendation based on electronic medical record and a medical-semantic-aware convolution neural network (MSCNN) for pediatric chronic cough.

BackgroundDue to the phenotypic similarities among different pediatric respiratory diseases with chronic cough, primary doctors often misdiagnose and the misuse of examinations is prevalent. In the pre-diagnosis stage, the patients' chief complaints and other information in the electronic medical record (EMR) provide a powerful reference for respiratory experts to make preliminary disease judgment and examination plan. In this paper, we proposed an intelligent prediagnosis system to predict disease diagnosis and recommend examinations based on EMR text.MethodsWe examined the clinical notes of 178,293 children with chronic cough symptoms from retrospective EMR data. The dataset is split into 7:3 for training and testing. From the testing set, we also extract 5% of samples for validation. We proposed a medical-semantic-aware convolution neural network (MSCNN) framework that can accomplish two downstream tasks from the same medical language model through transfer learning. First, a medical language model based on the word2vec algorithm was built to generate embeddings for the text data. Then, text convolutional neural network (TextCNN) was used to build models for disease prediction and examination recommendation.ResultsWe implemented 5 algorithms for disease prediction. In the disease prediction task, our algorithm outperformed the baseline methods on all metrics, with a top-1 accuracy (AC) of 0.68 and a top-3 AC of 0.923 on the testing set. By adding data enhancement, the top-3 AC reached 0.926. In the examination recommendation task, the overall AC on the testing set was 0.93 and the macro average (MA) F1-score was 0.88. The average area under the curve (AUC) on the training set was 0.97 while on the testing set it was 0.86.ConclusionsWe constructed an intelligent prediagnosis system with an MSCNN framework that can predict diseases and make examination recommendations based on EMR data. Our approach achieved good results on a retrospective clinical dataset and thus has great potential for the application of automated diagnosis assist in clinical practice during pre-diagnosis stage, which will provide help for primary level doctors or doctors in basic-level hospitals. Due to the generality of the proposed framework, it can be straight forwardly extended to prediagnosis for other diseases.

Prediction of tidal-driven currents using convolutional neural network

Efficient forecasting models of coastal currents are required for the predictive control of smart vessels with autonomous or remote operating capabilities. Accurate physics-based numerical models demand computational resources that often are unavailable onboard ships; moreover, internet bandwidth at remote locations might not be sufficient to exchange the large computational files with onshore servers. In this case, light and efficient Data-Driven Models (DDM) may be developed to be used as a surrogate for a physics-based model.Currents and surface elevation are hydrodynamic variables driven by astronomic tide potential, allowing for harmonic analysis and synthesis of otherwise irregular solutions of Navier-Stokes equations. Once numerical solutions are obtained and analysed in the Fourier domain, the identified harmonics (tide constituents) can be used as a dataset to train the DDM. One approach to building a DDM is by utilizing a Convolutional Neural Network (CNN).CNN is mainly applied in image recognition processes, where it learns the spatial features of image matrices of pixels arranged in columns and rows and uses the learned features to predict an output. This approach can be applied to 2D or 3D fields of predominantly tidal-driven coastal currents. This paper extends the CNN framework to approximate and predict coastal hydrodynamic state variables in space and time. The depth-averaged Navier-Stokes equations were used to create the dataset necessary to train the adopted CNN autoencoder architecture.Two methods of CNN model development are explored. The first method assesses CNN’s capabilities for time forward prediction, using 2D time series of hydrodynamic variables. The second method investigates CNN’s parameter (using Manning number) prediction ability using 2D spatial plots formed by the top 20 most energetic tidal constituents that were obtained from the numerical solution. The hyperparameters of the CNN model for both methods were carefully optimized and tuned using the loss function, Root Mean Square Error (RMSE). The coefficient of determination, the relative error of the time series, and its amplitude and phase were used to gauge the performance of the CNN model for both methods at the final testing stage.Once accurate solutions are achieved, the trained DDM is almost instantaneous and uses negligible computational resources as compared to traditional fluid solvers. Therefore, trained DDM can be deployed onboard ships where global forecast information (e.g., from Global Forecast System) can be used to provide an accurate prediction of the current forcing. To generate the instantaneous current predictions for navigation and smart ship control, the global data can be pulled or pushed to the ship before the mission.

Surrogate and inverse modeling for two-phase flow in porous media via theory-guided convolutional neural network

The theory-guided convolutional neural network (TgCNN) framework, which can incorporate discretized governing equation residuals into the training of convolutional neural networks (CNNs), is extended to two-phase porous media flow problems in this work. The two principal variables of the considered problem, pressure and saturation, are approximated simultaneously with two CNNs, respectively. Pressure and saturation are coupled with each other in the governing equations, and thus the two networks are also mutually conditioned in the training process by the discretized governing equations, which also increases the difficulty of model training. The coupled and discretized equations can provide valuable information in the training process. With the assistance of theory-guidance, the TgCNN surrogates can achieve better accuracy than ordinary CNN surrogates in two-phase flow problems. Moreover, a piecewise training strategy is proposed for the scenario with varying well controls, in which the TgCNN surrogates are constructed for different segments on the time dimension and stacked together to predict solutions for the whole time-span. For scenarios with larger variance of the formation property field, the TgCNN surrogates can also achieve satisfactory performance. The constructed TgCNN surrogates are further used for inversion of permeability fields by combining them with the iterative ensemble smoother (IES) algorithm, and sufficient inversion accuracy is obtained with improved efficiency.

A convolutional neural network method based on Adam optimizer with power-exponential learning rate for bearing fault diagnosis

The extraction of early fault features from time-series data is very crucial for convolutional neural networks (CNNs) in bearing fault diagnosis. To address this problem, a CNN framework based on identity mapping and Adam optimizer is presented for learning temporal dependencies and extracting fault features. The introduction of four identity mappings allows the deep layers to directly learn the data from the shallow layers, which alleviates the gradient disappearance problem caused by the increase of network depth. A new Adam optimizer with power-exponential learning rate is proposed to control the iteration direction and step size of CNN method, which solves the problems of local minima, overshoot or oscillation caused by the fixed values of the learning rates during the updating of network parameters. Compared to existed methods, the identification accuracy of the proposed method outperformed that of other methods for bearing fault diagnosis.

A study on several critical problems on arrhythmia detection using varying-dimensional electrocardiography

Objective. This work tries to provide answers to several critical questions on varying-dimensional electrocardiography (ECG) raised by the PhysioNet/Computing in Cardiology Challenge 2021 (CinC2021): can subsets of the standard 12 leads provide models with adequate information to give comparable performances for classifying ECG abnormalities? Can models be designed to be effective enough to classify a broad range of ECG abnormalities? Approach. To tackle these problems, we (challenge team name ‘Revenger’) propose several novel architectures within the framework of convolutional recurrent neural networks. These deep learning models are proven effective, and moreover, they provide comparable performances on reduced-lead ECGs, even in the extreme case of 2-lead ECGs. In addition, we propose a ‘lead-wise’ mechanism to facilitate parameter reuse of ECG neural network models. This mechanism largely reduces model sizes while keeping comparable performances. To further augment model performances on specific ECG abnormalities and to improve interpretability, we manually design auxiliary detectors based on clinical diagnostic rules. Main Results. In the post-challenge session, our approach achieved a challenge score of 0.38, 0.40, 0.41, 0.40, 0.35 on the 12, 6, 4, 3, 2-lead subsets respectively on the CinC2021 hidden test set. Significance. The proposed approach gives positive answers to the critical questions CinC2021 raises and lays a solid foundation for further research in the future on these topics.

Research on classification of benign and malignant lung nodules based on three-dimensional multi-view squeeze-and-excitation convolutional neural network

Lung cancer is the most threatening tumor disease to human health. Early detection is crucial to improve the survival rate and recovery rate of lung cancer patients. Existing methods use the two-dimensional multi-view framework to learn lung nodules features and simply integrate multi-view features to achieve the classification of benign and malignant lung nodules. However, these methods suffer from the problems of not capturing the spatial features effectively and ignoring the variability of multi-views. Therefore, this paper proposes a three-dimensional (3D) multi-view convolutional neural network (MVCNN) framework. To further solve the problem of different views in the multi-view model, a 3D multi-view squeeze-and-excitation convolution neural network (MVSECNN) model is constructed by introducing the squeeze-and-excitation (SE) module in the feature fusion stage. Finally, statistical methods are used to analyze model predictions and doctor annotations. In the independent test set, the classification accuracy and sensitivity of the model were 96.04% and 98.59% respectively, which were higher than other state-of-the-art methods. The consistency score between the predictions of the model and the pathological diagnosis results was 0.948, which is significantly higher than that between the doctor annotations and the pathological diagnosis results. The methods presented in this paper can effectively learn the spatial heterogeneity of lung nodules and solve the problem of multi-view differences. At the same time, the classification of benign and malignant lung nodules can be achieved, which is of great significance for assisting doctors in clinical diagnosis.

Smear character recognition method of side-end power meter based on PCA image enhancement

Abstract Since it is difficult for manual recording to track the rapid change of indication of the power meter, the power meter images are collected by the camera and automatically recognized and recorded to effectively overcome the disadvantages of manual recording. However, the complex scene lighting environment and smearing character shadows make it difficult to transfer captured images directly to convolutional neural networks for character recognition. A smear character recognition method of side-end power meter under complex lighting conditions is proposed in this article. First, the uneven illumination image enhancement algorithm is studied. Through the estimation of the illumination component of the image, the fusion weight is calculated by the principal component analysis for multiscale fusion, and the up-sampling and down-sampling are adopted to reduce the calculation of the algorithm and achieve the rapid image enhancement. A convolution neural network framework based on deep learning is proposed to realize the segmentation of smear characters, and the final segmented individual characters are fed into a network to identify the meter readings. The experimental results show that the proposed smear character recognition method has fast recognition speed, and the recognition rate of samples with the smear character and complex illumination reaches 99.8%, which meets the requirements of power meter character recognition and is better than other algorithms.

A Multiscale CNN Framework for Wireless Technique Classification in Internet of Things

Wireless technique classification (WTC) is of crucial importance in Internet of Things for realizing efficient spectrum sharing and interference management. However, the existing deep-learning-based methods have low classification accuracy, especially at low signal-to-noise ratio levels. In this article, a multiscale convolutional neural network framework is proposed for WTC. A multiscale module is exploited to capture the higher abstraction features. Simulation results demonstrate that our proposed scheme can achieve a better classification performance and a higher convergence speed compared to the state-of-the-art schemes.

Automatic Segmentation of MRI of Brain Tumor Using Deep Convolutional Network.

Computer-aided diagnosis and treatment of multimodal magnetic resonance imaging (MRI) brain tumor image segmentation has always been a hot and significant topic in the field of medical image processing. Multimodal MRI brain tumor image segmentation utilizes the characteristics of each modal in the MRI image to segment the entire tumor and tumor core area and enhanced them from normal brain tissues. However, the grayscale similarity between brain tissues in various MRI images is very immense making it difficult to deal with the segmentation of multimodal MRI brain tumor images through traditional algorithms. Therefore, we employ the deep learning method as a tool to make full use of the complementary feature information between the multimodalities and instigate the following research: (i) build a network model suitable for brain tumor segmentation tasks based on the fully convolutional neural network framework and (ii) adopting an end-to-end training method, using two-dimensional slices of MRI images as network input data. The problem of unbalanced categories in various brain tumor image data is overcome by introducing the Dice loss function into the network to calculate the network training loss; at the same time, parallel Dice loss is proposed to further improve the substructure segmentation effect. We proposed a cascaded network model based on a fully convolutional neural network to improve the tumor core area and enhance the segmentation accuracy of the tumor area and achieve good prediction results for the substructure segmentation on the BraTS 2017 data set.

Noise reduction technique using deep learning for ultrasound imaging during high-intensity focused ultrasound treatment

One of the problems with ultrasound imaging during high-intensity focused ultrasound (HIFU) treatment is that the therapeutic ultrasound components interfere with the diagnostic ultrasound components, making it impossible to monitor the tissue changes during HIFU exposure. In this study, a convolutional neural network (CNN) framework was applied to the reconstructed ultrasound images with HIFU noise to remove the therapeutic ultrasound components while the diagnostic ultrasound components remain intact. In the experiments, the chicken breast was used as a tissue sample and exposed to HIFU in the water tank. The ultrasound images with and without noise were acquired during an intermission period of HIFU exposure and the noise-reduced images was predicted using the proposed multi-layer regression CNN model through the training process. As a result, ultrasound images with sufficient spatial resolution to detect the thermal lesion were acquired.

DeephageTP: a convolutional neural network framework for identifying phage-specific proteins from metagenomic sequencing data.

Bacteriophages (phages) are the most abundant and diverse biological entity on Earth. Due to the lack of universal gene markers and database representatives, there about 50–90% of genes of phages are unable to assign functions. This makes it a challenge to identify phage genomes and annotate functions of phage genes efficiently by homology search on a large scale, especially for newly phages. Portal (portal protein), TerL (large terminase subunit protein), and TerS (small terminase subunit protein) are three specific proteins of Caudovirales phage. Here, we developed a CNN (convolutional neural network)-based framework, DeephageTP, to identify the three specific proteins from metagenomic data. The framework takes one-hot encoding data of original protein sequences as the input and automatically extracts predictive features in the process of modeling. To overcome the false positive problem, a cutoff-loss-value strategy is introduced based on the distributions of the loss values of protein sequences within the same category. The proposed model with a set of cutoff-loss-values demonstrates high performance in terms of Precision in identifying TerL and Portal sequences (94% and 90%, respectively) from the mimic metagenomic dataset. Finally, we tested the efficacy of the framework using three real metagenomic datasets, and the results shown that compared to the conventional alignment-based methods, our proposed framework had a particular advantage in identifying the novel phage-specific protein sequences of portal and TerL with remote homology to their counterparts in the training datasets. In summary, our study for the first time develops a CNN-based framework for identifying the phage-specific protein sequences with high complexity and low conservation, and this framework will help us find novel phages in metagenomic sequencing data. The DeephageTP is available at https://github.com/chuym726/DeephageTP.

Deep Learning-Based Prediction of Hematoma Expansion Using a Single Brain Computed Tomographic Slice in Patients With Spontaneous Intracerebral Hemorrhages

We aimed to predict hematoma expansion in intracerebral hemorrhage (ICH) patients by using the deep learning technique. We retrospectively collected data from ICH patients treated between May 2015 and May 2019. Head computed tomography (CT) scans were performed at admission, and 6 hours, 24 hours, and 72 hours after admission. CT scans were mandatory when neurologic deficits occurred. Univariate and multivariate analyses were conducted to illustrate the association between clinical variables and hematoma expansion. Convolutional neural network (CNN) was adopted to predict hematoma expansion based on brain CT slices. In addition, 5 machine learning methods, including support vector machine, multi-layer perceptron, naive Bayes, decision tree, and random forest, were also performed to predict hematoma expansion based on clinical variables for comparisons. A total of 223 patients were included. It was revealed that patients' older age (odds ratio [95% confidence interval]: 1.783 [1.417-1.924]), cerebral hemorrhage and breaking into the ventricle (2.524 [1.291-1.778]), coagulopathy (2.341 [1.677-3.454]), and baseline National Institutes of Health Stroke Scale (1.545 [1.132-3.203]) and Glasgow Coma Scale scores (0.782 [0.432-0.918]) independently associated with hematoma expanding. After 4-5 epochs, the CNN framework was well trained. The average sensitivity, specificity, and accuracy of CNN prediction are 0.9197, 0.8837, and 0.9058, respectively. Compared with 5 machine learning methods based on clinical variables, CNN can also achieve better performance. More than 90% of hematomas with or without expansion can be precisely classified by deep learning technology within this study, which is better than other methods based on clinical variables only. Deep learning technology could favorably predict hematoma expansion from non-contrast CT scan images.

Machine-Learning for Mapping and Monitoring Shallow Coral Reef Habitats

Mapping and monitoring coral reef benthic composition using remotely sensed imagery provides a large-scale inference of spatial and temporal dynamics. These maps have become essential components in marine science and management, with their utility being dependent upon accuracy, scale, and repeatability. One of the primary factors that affects the utility of a coral reef benthic composition map is the choice of the machine-learning algorithm used to classify the coral reef benthic classes. Current machine-learning algorithms used to map coral reef benthic composition and detect changes over time achieve moderate to high overall accuracies yet have not demonstrated spatio-temporal generalisation. The inability to generalise limits their scalability to only those reefs where in situ reference data samples are present. This limitation is becoming more pronounced given the rapid increase in the availability of high temporal (daily) and high spatial resolution (<5 m) multispectral satellite imagery. Therefore, there is presently a need to identify algorithms capable of spatio-temporal generalisation in order to increase the scalability of coral reef benthic composition mapping and change detection. This review focuses on the most commonly used machine-learning algorithms applied to map coral reef benthic composition and detect benthic changes over time using multispectral satellite imagery. The review then introduces convolutional neural networks that have recently demonstrated an ability to spatially and temporally generalise in relation to coral reef benthic mapping; and recurrent neural networks that have demonstrated spatio-temporal generalisation in the field of land cover change detection. A clear conclusion of this review is that existing convolutional neural network and recurrent neural network frameworks hold the most potential in relation to increasing the spatio-temporal scalability of coral reef benthic composition mapping and change detection due to their ability to spatially and temporally generalise.

A lightweight 3D-2D convolutional neural network for spectral-spatial classification of hyperspectral images

Hyperspectral Image (HSI) is usually composed of hundreds of capturing wavelength bands, which not only increase the size of the HSI rapidly but also impose various obstacles in classifying the objects accurately. Moreover, the traditional machine learning schemes utilize only the spectral features for HSI classification, which, therefore, neglect the spatial features that have a significant impact on the classification improvement. To address the aforementioned issues, in this paper, we propose to employ the principal component analysis (PCA), the baseline feature extraction method, and a thoughtfully designed stacked autoencoder, a deep learning-based feature extraction approach, for reducing the high dimensionality of the HSI and then propose a novel lightweight 3D-2D convolutional neural network (CNN) framework to concurrently exploit both spatial and spectral features from the dimensionality-reduced HSI for classification. In particular, PCA and stacked autoencoder are applied to reduce the high dimensionality of the original HSI and then the proposed 3D-2D CNN provides a combination of 3D and 2D convolution operations to extract the subtle spatial and spectral features for efficient classification. We well-adjust the proposed 3D-2D CNN architecture, and perform extensive experiments on three benchmark HSI datasets and compare our approach with the state-of-the-art classical and deep learning methods. Experimental results illustrate that we have achieved an overall accuracy of 99.73%, 99.90%, and 99.32% on Indian Pines, Pavia University, and Kennedy Space Center datasets, respectively, which outperform the classical machine learning and independent 2D and 3D CNN-based state-of-the-art methods.

A Global Appearance and Local Coding Distortion Based Fusion Framework for CNN Based Filtering in Video Coding

In-loop filtering is used in video coding to process the reconstructed frame in order to remove blocking artifacts. With the development of convolutional neural networks (CNNs), CNNs have been explored for in-loop filtering considering it can be treated as an image de-noising task. However, in addition to being a distorted image, the reconstructed frame is also obtained by a fixed line of block based encoding operations in video coding. It carries coding-unit based coding distortion of some similar characteristics. Therefore, in this paper, we address the filtering problem from two aspects, (i) global appearance restoration for disrupted texture and (ii) local coding distortion restoration caused by fixed pipeline of coding. Accordingly, a three-stream global appearance and local coding distortion based fusion network is developed with a high-level global feature stream, a high-level local feature stream and a low-level local feature stream. Ablation study is conducted to validate the necessity of different features, demonstrating that the global features and local features can complement each other in filtering and achieve better performance when combined. To the best of our knowledge, we are the first one that clearly characterizes the video filtering process from the above global appearance and local coding distortion restoration aspects with experimental verification, providing a clear pathway to developing filter techniques. Experimental results demonstrate that the proposed method significantly outperforms the existing single-frame based methods and achieves 13.5%, 11.3%, 11.7% BD-Rate saving on average for AI, LDP and RA configurations, respectively, compared with the HEVC reference software.

Automatic lesion segmentation using atrous convolutional deep neural networks in dermoscopic skin cancer images

BackgroundMelanoma is the most dangerous and aggressive form among skin cancers, exhibiting a high mortality rate worldwide. Biopsy and histopathological analysis are standard procedures for skin cancer detection and prevention in clinical settings. A significant step in the diagnosis process is the deep understanding of the patterns, size, color, and structure of lesions based on images obtained through dermatoscopes for the infected area. However, the manual segmentation of the lesion region is time-consuming because the lesion evolves and changes its shape over time, making its prediction challenging. Moreover, it is challenging to predict melanoma at the initial stage as it closely resembles other skin cancer types that are not malignant as melanoma; thus, automatic segmentation techniques are required to design a computer-aided system for accurate and timely detection.MethodsAs deep learning approaches have gained significant attention in recent years due to their remarkable performance, therefore, in this work, we proposed a novel design of a convolutional neural network (CNN) framework based on atrous convolutions for automatic lesion segmentation. This architecture is built based on the concept of atrous/dilated convolutions which are effective for semantic segmentation. A deep neural network is designed from scratch employing several building blocks consisting of convolutional, batch normalization, leakyReLU layer, and fine-tuned hyperparameters contributing altogether towards higher performance.ConclusionThe network was tested on three benchmark datasets provided by International Skin Imaging Collaboration (ISIC), i.e., ISIC 2016, ISIC 2017, and ISIC 2018. The experimental results showed that the proposed network achieved an average Jaccard index of 90.4% on ISIC 2016, 81.8% on ISIC 2017, and 89.1% on ISIC 2018 datasets, respectively which is recorded as higher than the top three winners of the ISIC challenge and other state-of-the-art methods. Also, the model successfully extracts lesions from the whole image in one pass in less time, requiring no pre-processing step. The conclusions yielded that network is accurate in performing lesion segmentation on adopted datasets.

End-to-End Convolutional Neural Network Framework for Breast Ultrasound Analysis Using Multiple Parametric Images Generated from Radiofrequency Signals

Breast ultrasound (BUS) is an effective clinical modality for diagnosing breast abnormalities in women. Deep-learning techniques based on convolutional neural networks (CNN) have been widely used to analyze BUS images. However, the low quality of B-mode images owing to speckle noise and a lack of training datasets makes BUS analysis challenging in clinical applications. In this study, we proposed an end-to-end CNN framework for BUS analysis using multiple parametric images generated from radiofrequency (RF) signals. The entropy and phase images, which represent the microstructural and anatomical information, respectively, and the traditional B-mode images were used as parametric images in the time domain. In addition, the attenuation image, estimated from the frequency domain using RF signals, was used for the spectral features. Because one set of RF signals from one patient produced multiple images as CNN inputs, the proposed framework overcame the limitation of datasets in a broad sense of data augmentation while providing complementary information to compensate for the low quality of the B-mode images. The experimental results showed that the proposed architecture improved the classification accuracy and recall by 5.5% and 11.6%, respectively, compared with the traditional approach using only B-mode images. The proposed framework can be extended to various other parametric images in both the time and frequency domains using deep neural networks to improve its performance.

Error-Correcting Output Codes in the Framework of Deep Ordinal Classification

Automatic classification tasks on structured data have been revolutionized by Convolutional Neural Networks (CNNs), but the focus has been on binary and nominal classification tasks. Only recently, ordinal classification (where class labels present a natural ordering) has been tackled through the framework of CNNs. Also, ordinal classification datasets commonly present a high imbalance in the number of samples of each class, making it an even harder problem. Focus should be shifted from classic classification metrics towards per-class metrics (like AUC or Sensitivity) and rank agreement metrics (like Cohen’s Kappa or Spearman’s rank correlation coefficient). We present a new CNN architecture based on the Ordinal Binary Decomposition (OBD) technique using Error-Correcting Output Codes (ECOC). We aim to show experimentally, using four different CNN architectures and two ordinal classification datasets, that the OBD+ECOC methodology significantly improves the mean results on the relevant ordinal and class-balancing metrics. The proposed method is able to outperform a nominal approach as well as already existing ordinal approaches, achieving a mean performance of {{,mathrm{textit{RMSE}},}}= 1.0797 for the Retinopathy dataset and {{,mathrm{textit{RMSE}},}}= 1.1237 for the Adience dataset averaged over 4 different architectures.

Automatic Detection and Segmentation of Thrombi in Abdominal Aortic Aneurysms Using a Mask Region-Based Convolutional Neural Network with Optimized Loss Functions.

The detection and segmentation of thrombi are essential for monitoring the disease progression of abdominal aortic aneurysms (AAAs) and for patient care and management. As they have inherent capabilities to learn complex features, deep convolutional neural networks (CNNs) have been recently introduced to improve thrombus detection and segmentation. However, investigations into the use of CNN methods is in the early stages and most of the existing methods are heavily concerned with the segmentation of thrombi, which only works after they have been detected. In this work, we propose a fully automated method for the whole process of the detection and segmentation of thrombi, which is based on a well-established mask region-based convolutional neural network (Mask R-CNN) framework that we improve with optimized loss functions. The combined use of complete intersection over union (CIoU) and smooth L1 loss was designed for accurate thrombus detection and then thrombus segmentation was improved with a modified focal loss. We evaluated our method against 60 clinically approved patient studies (i.e., computed tomography angiography (CTA) image volume data) by conducting 4-fold cross-validation. The results of comparisons to multiple other state-of-the-art methods suggested the superior performance of our method, which achieved the highest F1 score for thrombus detection (0.9197) and outperformed most metrics for thrombus segmentation.