A novel stacked generalization of models for improved TB detection in chest radiographs

Lung cancer detection with machine learning classifiers with multi-attribute decision-making system and deep learning model

Diseases of the airways and the other parts of the lung cause chronic respiratory diseases. The major cause of lung disease is tobacco smoke, along with risk factors such as dust, air pollution, chemicals, and frequent lower respiratory infections during childhood. Early detection of these diseases requires the analysis of medical images, which would aid doctors in providing effective treatment.This paper aims to classify lung X-ray images as benign or malignant and to identify the type of disease, such as Atelectasis, Infiltration, Nodule, and Pneumonia, if the disease is malignant. Machine learning (ML) approaches, combined with a multi-attribute decision-making method called Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), are used to rank different classifiers. Additionally, the deep learning (DL) model Inception v3 is proposed. This method ranks the SVM with RBF as the best classifier among the others used in this approach. Furthermore, the results show that the deep learning model achieves the best accuracy of 97.05%, which is 11.8% higher than the machine learning approach using the same dataset.

Diagnostic Performance of Artificial Intelligence-Based Methods for Tuberculosis Detection: Systematic Review

BACKGROUND

Tuberculosis (TB) remains a significant health concern, contributing to the highest mortality among infectious diseases worldwide. However, none of the various TB diagnostic tools introduced is deemed sufficient on its own for the diagnostic pathway, so various artificial intelligence (AI)-based methods have been developed to address this issue.

OBJECTIVE

We aimed to provide a comprehensive evaluation of AI-based algorithms for TB detection across various data modalities.

METHODS

Following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) 2020 guidelines, we conducted a systematic review to synthesize current knowledge on this topic. Our search across 3 major databases (Scopus, PubMed, Association for Computing Machinery [ACM] Digital Library) yielded 1146 records, of which we included 152 (13.3%) studies in our analysis. QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies version 2) was performed for the risk-of-bias assessment of all included studies.

RESULTS

Radiographic biomarkers (n=129, 84.9%) and deep learning (DL; n=122, 80.3%) approaches were predominantly used, with convolutional neural networks (CNNs) using Visual Geometry Group (VGG)-16 (n=37, 24.3%), ResNet-50 (n=33, 21.7%), and DenseNet-121 (n=19, 12.5%) architectures being the most common DL approach. The majority of studies focused on model development (n=143, 94.1%) and used a single modality approach (n=141, 92.8%). AI methods demonstrated good performance in all studies: mean accuracy=91.93% (SD 8.10%, 95% CI 90.52%-93.33%; median 93.59%, IQR 88.33%-98.32%), mean area under the curve (AUC)=93.48% (SD 7.51%, 95% CI 91.90%-95.06%; median 95.28%, IQR 91%-99%), mean sensitivity=92.77% (SD 7.48%, 95% CI 91.38%-94.15%; median 94.05% IQR 89%-98.87%), and mean specificity=92.39% (SD 9.4%, 95% CI 90.30%-94.49%; median 95.38%, IQR 89.42%-99.19%). AI performance across different biomarker types showed mean accuracies of 92.45% (SD 7.83%), 89.03% (SD 8.49%), and 84.21% (SD 0%); mean AUCs of 94.47% (SD 7.32%), 88.45% (SD 8.33%), and 88.61% (SD 5.9%); mean sensitivities of 93.8% (SD 6.27%), 88.41% (SD 10.24%), and 93% (SD 0%); and mean specificities of 94.2% (SD 6.63%), 85.89% (SD 14.66%), and 95% (SD 0%) for radiographic, molecular/biochemical, and physiological types, respectively. AI performance across various reference standards showed mean accuracies of 91.44% (SD 7.3%), 93.16% (SD 6.44%), and 88.98% (SD 9.77%); mean AUCs of 90.95% (SD 7.58%), 94.89% (SD 5.18%), and 92.61% (SD 6.01%); mean sensitivities of 91.76% (SD 7.02%), 93.73% (SD 6.67%), and 91.34% (SD 7.71%); and mean specificities of 86.56% (SD 12.8%), 93.69% (SD 8.45%), and 92.7% (SD 6.54%) for bacteriological, human reader, and combined reference standards, respectively. The transfer learning (TL) approach showed increasing popularity (n=89, 58.6%). Notably, only 1 (0.7%) study conducted domain-shift analysis for TB detection.

CONCLUSIONS

Findings from this review underscore the considerable promise of AI-based methods in the realm of TB detection. Future research endeavors should prioritize conducting domain-shift analyses to better simulate real-world scenarios in TB detection.

TRIAL REGISTRATION

PROSPERO CRD42023453611; https://www.crd.york.ac.uk/PROSPERO/view/CRD42023453611.

A methodical exploration of imaging modalities from dataset to detection through machine learning paradigms in prominent lung disease diagnosis: a review

BACKGROUND

Lung diseases, both infectious and non-infectious, are the most prevalent cause of mortality overall in the world. Medical research has identified pneumonia, lung cancer, and Corona Virus Disease 2019 (COVID-19) as prominent lung diseases prioritized over others. Imaging modalities, including X-rays, computer tomography (CT) scans, magnetic resonance imaging (MRIs), positron emission tomography (PET) scans, and others, are primarily employed in medical assessments because they provide computed data that can be utilized as input datasets for computer-assisted diagnostic systems. Imaging datasets are used to develop and evaluate machine learning (ML) methods to analyze and predict prominent lung diseases.

OBJECTIVE

This review analyzes ML paradigms, imaging modalities' utilization, and recent developments for prominent lung diseases. Furthermore, the research also explores various datasets available publically that are being used for prominent lung diseases.

METHODS

The well-known databases of academic studies that have been subjected to peer review, namely ScienceDirect, arXiv, IEEE Xplore, MDPI, and many more, were used for the search of relevant articles. Applied keywords and combinations used to search procedures with primary considerations for review, such as pneumonia, lung cancer, COVID-19, various imaging modalities, ML, convolutional neural networks (CNNs), transfer learning, and ensemble learning.

RESULTS

This research finding indicates that X-ray datasets are preferred for detecting pneumonia, while CT scan datasets are predominantly favored for detecting lung cancer. Furthermore, in COVID-19 detection, X-ray datasets are prioritized over CT scan datasets. The analysis reveals that X-rays and CT scans have surpassed all other imaging techniques. It has been observed that using CNNs yields a high degree of accuracy and practicability in identifying prominent lung diseases. Transfer learning and ensemble learning are complementary techniques to CNNs to facilitate analysis. Furthermore, accuracy is the most favored metric for assessment.

Observer Performance Evaluation of a Deep Learning Model for Multilabel Classification of Active Tuberculosis Lung Zone-Wise Manifestations

Background Chest X-rays (CXRs) are widely used for cost-effective screening of active pulmonary tuberculosis despite their limitations in sensitivity and specificity when interpreted by clinicians or radiologists. To address this issue, computer-aided detection (CAD) algorithms, particularly deep learning architectures based on convolution, have been developed to automate the analysis of radiography imaging. Deep learning algorithms have shown promise in accurately classifying lung abnormalities using chest X-ray images. In this study, we utilized the EfficientNet B4 model, which was pre-trained on ImageNet with 380x380 input dimensions, using its weights for transfer learning, and was modified with a series of components including global average pooling, batch normalization, dropout, and a classifier with 12 image-wise and 44 segment-wise lung zone evaluation classes using sigmoid activation. Objectives Assess the clinical usefulness of our previously created EfficientNet B4 model in identifying lung zone-specific abnormalities related to active tuberculosis through an observer performance test involving a skilled clinician operating in tuberculosis-specific environments. Methods The ground truth was established by a radiologist who examined all sample CXRs to identify lung zone-wise abnormalities. An expert clinician working in tuberculosis-specific settings independently reviewed the same CXR with blinded access to the ground truth. Simultaneously, the CXRs were classified using the EfficientNet B4 model. The clinician's assessments were then compared with the model's predictions, and the agreement between the two was measured using the kappa coefficient, evaluating the model's performance in classifying active tuberculosis manifestations across lung zones. Results The results show a strong agreement (Kappa ≥0.81) seen for lung zone-wise abnormalities of pneumothorax, mediastinal shift, emphysema, fibrosis, calcifications, pleural effusion, and cavity. Substantial agreement (Kappa = 0.61-0.80) for cavity, mediastinal shift, volume loss, and collapsed lungs. The Kappa score for lung zone-wise abnormalities is moderate (0.41-0.60) for 39% of cases. In image-wise agreement, the EfficientNet B4 model's performance ranges from moderate to almost perfect across categories, while in lung zone-wise agreement, it varies from fair to almost perfect. The results show strong agreement between the EfficientNet B4 model and the human reader in detecting lung zone-wise and image-wise manifestations. Conclusion The clinical utility of the EfficientNet B4 models to detect the abnormalities can aid clinicians in primary care settings for screening and triaging tuberculosis where resources are constrained or overburdened.

Review on chest pathogies detection systems using deep learning techniques

Chest radiography is the standard and most affordable way to diagnose, analyze, and examine different thoracic and chest diseases. Typically, the radiograph is examined by an expert radiologist or physician to decide about a particular anomaly, if exists. Moreover, computer-aided methods are used to assist radiologists and make the analysis process accurate, fast, and more automated. A tremendous improvement in automatic chest pathologies detection and analysis can be observed with the emergence of deep learning. The survey aims to review, technically evaluate, and synthesize the different computer-aided chest pathologies detection systems. The state-of-the-art of single and multi-pathologies detection systems, which are published in the last five years, are thoroughly discussed. The taxonomy of image acquisition, dataset preprocessing, feature extraction, and deep learning models are presented. The mathematical concepts related to feature extraction model architectures are discussed. Moreover, the different articles are compared based on their contributions, datasets, methods used, and the results achieved. The article ends with the main findings, current trends, challenges, and future recommendations.

A Deep Learning-Based Framework for Uncertainty Quantification in Medical Imaging Using the DropWeak Technique: An Empirical Study with Baresnet

Lung cancer is a leading cause of cancer-related deaths globally. Early detection is crucial for improving patient survival rates. Deep learning (DL) has shown promise in the medical field, but its accuracy must be evaluated, particularly in the context of lung cancer classification. In this study, we conducted uncertainty analysis on various frequently used DL architectures, including Baresnet, to assess the uncertainties in the classification results. This study focuses on the use of deep learning for the classification of lung cancer, which is a critical aspect of improving patient survival rates. The study evaluates the accuracy of various deep learning architectures, including Baresnet, and incorporates uncertainty quantification to assess the level of uncertainty in the classification results. The study presents a novel automatic tumor classification system for lung cancer based on CT images, which achieves a classification accuracy of 97.19% with an uncertainty quantification. The results demonstrate the potential of deep learning in lung cancer classification and highlight the importance of uncertainty quantification in improving the accuracy of classification results. This study's novelty lies in the incorporation of uncertainty quantification in deep learning for lung cancer classification, which can lead to more reliable and accurate diagnoses in clinical settings.

Deep learning classification of active tuberculosis lung zones wise manifestations using chest X-rays: a multi label approach

Chest X-rays are the most economically viable diagnostic imaging test for active pulmonary tuberculosis screening despite the high sensitivity and low specificity when interpreted by clinicians or radiologists. Computer aided detection (CAD) algorithms, especially convolution based deep learning architecture, have been proposed to facilitate the automation of radiography imaging modalities. Deep learning algorithms have found success in classifying various abnormalities in lung using chest X-ray. We fine-tuned, validated and tested EfficientNetB4 architecture and utilized the transfer learning methodology for multilabel approach to detect lung zone wise and image wise manifestations of active pulmonary tuberculosis using chest X-ray. We used Area Under Receiver Operating Characteristic (AUC), sensitivity and specificity along with 95% confidence interval as model evaluation metrics. We also utilized the visualisation capabilities of convolutional neural networks (CNN), Gradient-weighted Class Activation Mapping (Grad-CAM) as post-hoc attention method to investigate the model and visualisation of Tuberculosis abnormalities and discuss them from radiological perspectives. EfficientNetB4 trained network achieved remarkable AUC, sensitivity and specificity of various pulmonary tuberculosis manifestations in intramural test set and external test set from different geographical region. The grad-CAM visualisations and their ability to localize the abnormalities can aid the clinicians at primary care settings for screening and triaging of tuberculosis where resources are constrained or overburdened.

A Survey on AI Techniques for Thoracic Diseases Diagnosis Using Medical Images

Thoracic diseases refer to disorders that affect the lungs, heart, and other parts of the rib cage, such as pneumonia, novel coronavirus disease (COVID-19), tuberculosis, cardiomegaly, and fracture. Millions of people die every year from thoracic diseases. Therefore, early detection of these diseases is essential and can save many lives. Earlier, only highly experienced radiologists examined thoracic diseases, but recent developments in image processing and deep learning techniques are opening the door for the automated detection of these diseases. In this paper, we present a comprehensive review including: types of thoracic diseases; examination types of thoracic images; image pre-processing; models of deep learning applied to the detection of thoracic diseases (e.g., pneumonia, COVID-19, edema, fibrosis, tuberculosis, chronic obstructive pulmonary disease (COPD), and lung cancer); transfer learning background knowledge; ensemble learning; and future initiatives for improving the efficacy of deep learning models in applications that detect thoracic diseases. Through this survey paper, researchers may be able to gain an overall and systematic knowledge of deep learning applications in medical thoracic images. The review investigates a performance comparison of various models and a comparison of various datasets.

Ensemble Technique Coupled with Deep Transfer Learning Framework for Automatic Detection of Tuberculosis from Chest X-ray Radiographs

Tuberculosis (TB) is an infectious disease affecting humans' lungs and is currently ranked the 13th leading cause of death globally. Due to advancements in technology and the availability of medical datasets, automatic analysis and classification of chest X-rays (CXRs) into TB and non-TB can be a reliable alternative for early TB screening. We propose an automatic TB detection system using advanced deep learning (DL) models. A substantial part of a CXR image is dark, with no relevant information for diagnosis and potentially confusing DL models. In this work, the U-Net model extracts the region of interest from CXRs and the segmented images are fed to the DL models for feature extraction. Eight different convolutional neural networks (CNN) models are employed in our experiments, and their classification performance is compared based on three publicly available CXR datasets. The U-Net model achieves segmentation accuracy of 98.58%, intersection over union (IoU) of 93.10, and a Dice coefficient score of 96.50. Our proposed stacked ensemble algorithm performed better by achieving accuracy, sensitivity, and specificity values of 98.38%, 98.89%, and 98.70%, respectively. Experimental results confirm that segmented lung CXR images with ensemble learning produce a better result than un-segmented lung CXR images.

Detection and Visualisation of Pneumoconiosis Using an Ensemble of Multi-Dimensional Deep Features Learned from Chest X-rays

Pneumoconiosis is a group of occupational lung diseases induced by mineral dust inhalation and subsequent lung tissue reactions. It can eventually cause irreparable lung damage, as well as gradual and permanent physical impairments. It has affected millions of workers in hazardous industries throughout the world, and it is a leading cause of occupational death. It is difficult to diagnose early pneumoconiosis because of the low sensitivity of chest radiographs, the wide variation in interpretation between and among readers, and the scarcity of B-readers, which all add to the difficulty in diagnosing these occupational illnesses. In recent years, deep machine learning algorithms have been extremely successful at classifying and localising abnormality of medical images. In this study, we proposed an ensemble learning approach to improve pneumoconiosis detection in chest X-rays (CXRs) using nine machine learning classifiers and multi-dimensional deep features extracted using CheXNet-121 architecture. There were eight evaluation metrics utilised for each high-level feature set of the associated cross-validation datasets in order to compare the ensemble performance and state-of-the-art techniques from the literature that used the same cross-validation datasets. It is observed that integrated ensemble learning exhibits promising results (92.68% accuracy, 85.66% Matthews correlation coefficient (MCC), and 0.9302 area under the precision-recall (PR) curve), compared to individual CheXNet-121 and other state-of-the-art techniques. Finally, Grad-CAM was used to visualise the learned behaviour of individual dense blocks within CheXNet-121 and their ensembles into three-color channels of CXRs. We compared the Grad-CAM-indicated ROI to the ground-truth ROI using the intersection of the union (IOU) and average-precision (AP) values for each classifier and their ensemble. Through the visualisation of the Grad-CAM within the blue channel, the average IOU passed more than 90% of the pneumoconiosis detection in chest radiographs.

Deep Learning Approaches for Automatic Localization in Medical Images

Recent revolutionary advances in deep learning (DL) have fueled several breakthrough achievements in various complicated computer vision tasks. The remarkable successes and achievements started in 2012 when deep learning neural networks (DNNs) outperformed the shallow machine learning models on a number of significant benchmarks. Significant advances were made in computer vision by conducting very complex image interpretation tasks with outstanding accuracy. These achievements have shown great promise in a wide variety of fields, especially in medical image analysis by creating opportunities to diagnose and treat diseases earlier. In recent years, the application of the DNN for object localization has gained the attention of researchers due to its success over conventional methods, especially in object localization. As this has become a very broad and rapidly growing field, this study presents a short review of DNN implementation for medical images and validates its efficacy on benchmarks. This study presents the first review that focuses on object localization using the DNN in medical images. The key aim of this study was to summarize the recent studies based on the DNN for medical image localization and to highlight the research gaps that can provide worthwhile ideas to shape future research related to object localization tasks. It starts with an overview on the importance of medical image analysis and existing technology in this space. The discussion then proceeds to the dominant DNN utilized in the current literature. Finally, we conclude by discussing the challenges associated with the application of the DNN for medical image localization which can drive further studies in identifying potential future developments in the relevant field of study.

Enhancing the weighted voting ensemble algorithm for tuberculosis predictive diagnosis

Tuberculosis has the most considerable death rate among diseases caused by a single micro-organism type. The disease is a significant issue for most third-world countries due to poor diagnosis and treatment potentials. Early diagnosis of tuberculosis is the most effective way of managing the disease in patients to reduce the mortality rate of the infection. Despite several methods that exist in diagnosing tuberculosis, the limitations ranging from the cost in carrying out the test to the time taken to obtain the results have hindered early diagnosis of the disease. This work aims to develop a predictive model that would help in the diagnosis of TB using an extended weighted voting ensemble method. The method used to carry out this research involved analyzing tuberculosis gene expression data obtained from GEO (Transcript Expression Omnibus) database and developing a classification model to aid tuberculosis diagnosis. A classifier combination of Naïve Bayes (NB), and Support Vector Machine (SVM) was used to develop the classification model. The weighted voting ensemble technique was used to improve the classification model's performance by combining the classification results of the single classifier and selecting the group with the highest vote based on the weights given to the single classifiers. Experimental analysis indicates a performance accuracy of the enhanced ensemble classifier as 0.95, which showed a better performance than the single classifiers, which had 0.92, and 0.87 obtained from SVM and NB, respectively. The developed model can also assist health practitioners in the timely diagnosis of tuberculosis, which would reduce the mortality rate caused by the disease, especially in developing countries.

Convolutional neural network-based ensemble methods to recognize Bangla handwritten character

In this era of advancements in deep learning, an autonomous system that recognizes handwritten characters and texts can be eventually integrated with the software to provide better user experience. Like other languages, Bangla handwritten text extraction also has various applications such as post-office automation, signboard recognition, and many more. A large-scale and efficient isolated Bangla handwritten character classifier can be the first building block to create such a system. This study aims to classify the handwritten Bangla characters. The proposed methods of this study are divided into three phases. In the first phase, seven convolutional neural networks i.e., CNN-based architectures are created. After that, the best performing CNN model is identified, and it is used as a feature extractor. Classifiers are then obtained by using shallow machine learning algorithms. In the last phase, five ensemble methods have been used to achieve better performance in the classification task. To systematically assess the outcomes of this study, a comparative analysis of the performances has also been carried out. Among all the methods, the stacked generalization ensemble method has achieved better performance than the other implemented methods. It has obtained accuracy, precision, and recall of 98.68%, 98.69%, and 98.68%, respectively on the Ekush dataset. Moreover, the use of CNN architectures and ensemble methods in large-scale Bangla handwritten character recognition has also been justified by obtaining consistent results on the BanglaLekha-Isolated dataset. Such efficient systems can move the handwritten recognition to the next level so that the handwriting can easily be automated.

A voting-based ensemble deep learning method focusing on image augmentation and preprocessing variations for tuberculosis detection

Tuberculosis (TB) is known as a potentially dangerous and infectious disease that affects mostly lungs worldwide. The detection and treatment of TB at an early stage are critical for preventing the disease and decreasing the risk of mortality and transmission of it to others. Nowadays, as the most common medical imaging technique, chest radiography (CXR) is useful for determining thoracic diseases. Computer-aided detection (CADe) systems are also crucial mechanisms to provide more reliable, efficient, and systematic approaches with accelerating the decision-making process of clinicians. In this study, we propose voting and preprocessing variations-based ensemble CNN model for TB detection. We utilize 40 different variations in fine-tuned CNN models based on InceptionV3 and Xception by also using CLAHE (contrast-limited adaptive histogram equalization) preprocessing technique and 10 different image transformations for data augmentation types. After analyzing all these combination schemes, three or five best classifier models are selected as base learners for voting operations. We apply the Bayesian optimization-based weighted voting and the average of probabilities as a combination rule in soft voting methods on two TB CXR image datasets to get better results in various numbers of models. The computational results indicate that the proposed method achieves 97.500% and 97.699% accuracy rates on Montgomery and Shenzhen datasets, respectively. Furthermore, our method outperforms state-of-the-art results for the two TB detection datasets in terms of accuracy rate.

Novel Deep Learning Technique Used in Management and Discharge of Hospitalized Patients with COVID-19 in China

Purpose

The low sensitivity and false-negative results of nucleic acid testing greatly affect its performance in diagnosing and discharging patients with coronavirus disease (COVID-19). Chest computed tomography (CT)-based evaluation of pneumonia may indicate a need for isolation. Therefore, this radiologic modality plays an important role in managing patients with suspected COVID-19. Meanwhile, deep learning (DL) technology has been successful in detecting various imaging features of chest CT. This study applied a novel DL technique to standardize the discharge criteria of COVID-19 patients with consecutive negative respiratory pathogen nucleic acid test results at a “square cabin” hospital.

Patients and Methods

DL was used to evaluate the chest CT scans of 270 hospitalized COVID-19 patients who had two consecutive negative nucleic acid tests (sampling interval >1 day). The CT scans evaluated were obtained after the patients’ second negative test result. The standard criterion determined by DL for patient discharge was a total volume ratio of lesion to lung <50%.

Results

The mean number of days between hospitalization and DL was 14.3 (± 2.4). The average intersection over union was 0.7894. Two hundred and thirteen (78.9%) patients exhibited pneumonia, of whom 54.0% (115/213) had mild interstitial fibrosis. Twenty-one, 33, and 4 cases exhibited vascular enlargement, pleural thickening, and mediastinal lymphadenopathy, respectively. Of the latter, 18.8% (40/213) had a total volume ratio of lesions to lung ≥50% according to our severity scale and were monitored continuously in the hospital. Three cases had a positive follow-up nucleic acid test during hospitalization. None of the 230 discharged cases later tested positive or exhibited pneumonia progression.

Conclusion

The novel DL enables the accurate management of hospitalized patients with COVID-19 and can help avoid cluster transmission or exacerbation in patients with false-negative acid test.

A Survey of Deep Learning for Lung Disease Detection on Medical Images: State-of-the-Art, Taxonomy, Issues and Future Directions

The recent developments of deep learning support the identification and classification of lung diseases in medical images. Hence, numerous work on the detection of lung disease using deep learning can be found in the literature. This paper presents a survey of deep learning for lung disease detection in medical images. There has only been one survey paper published in the last five years regarding deep learning directed at lung diseases detection. However, their survey is lacking in the presentation of taxonomy and analysis of the trend of recent work. The objectives of this paper are to present a taxonomy of the state-of-the-art deep learning based lung disease detection systems, visualise the trends of recent work on the domain and identify the remaining issues and potential future directions in this domain. Ninety-eight articles published from 2016 to 2020 were considered in this survey. The taxonomy consists of seven attributes that are common in the surveyed articles: image types, features, data augmentation, types of deep learning algorithms, transfer learning, the ensemble of classifiers and types of lung diseases. The presented taxonomy could be used by other researchers to plan their research contributions and activities. The potential future direction suggested could further improve the efficiency and increase the number of deep learning aided lung disease detection applications.

Assessment of an ensemble of machine learning models toward abnormality detection in chest radiographs

Respiratory diseases account for a significant proportion of deaths and disabilities across the world. Chest X-ray (CXR) analysis remains a common diagnostic imaging modality for confirming intra-thoracic cardiopulmonary abnormalities. However, there remains an acute shortage of expert radiologists, particularly in under-resourced settings, resulting in severe interpretation delays. These issues can be mitigated by a computer-aided diagnostic (CADx) system to supplement decision-making and improve throughput while preserving and possibly improving the standard-of-care. Systems reported in the literature or popular media use handcrafted features and/or data-driven algorithms like deep learning (DL) to learn underlying data distributions. The remarkable success of convolutional neural networks (CNN) toward image recognition tasks has made them a promising choice for automated medical image analyses. However, CNNs suffer from high variance and may overfit due to their sensitivity to training data fluctuations. Ensemble learning helps to reduce this variance by combining predictions of multiple learning algorithms to construct complex, non-linear functions and improve robustness and generalization. This study aims to construct and assess the performance of an ensemble of machine learning (ML) models applied to the challenge of classifying normal and abnormal CXRs and significantly reducing the diagnostic load of radiologists and primary-care physicians.

Ensemble Deep Learning for Diabetic Retinopathy Detection Using Optical Coherence Tomography Angiography

Purpose

To evaluate the role of ensemble learning techniques with deep learning in classifying diabetic retinopathy (DR) in optical coherence tomography angiography (OCTA) images and their corresponding co-registered structural images.

Methods

A total of 463 volumes from 380 eyes were acquired using the 3 × 3-mm OCTA protocol on the Zeiss Plex Elite system. Enface images of the superficial and deep capillary plexus were exported from both the optical coherence tomography and OCTA data. Component neural networks were constructed using single data-types and fine-tuned using VGG19, ResNet50, and DenseNet architectures pretrained on ImageNet weights. These networks were then ensembled using majority soft voting and stacking techniques. Results were compared with a classifier using manually engineered features. Class activation maps (CAMs) were created using the original CAM algorithm and Grad-CAM.

Results

The networks trained with the VGG19 architecture outperformed the networks trained on deeper architectures. Ensemble networks constructed using the four fine-tuned VGG19 architectures achieved accuracies of 0.92 and 0.90 for the majority soft voting and stacking methods respectively. Both ensemble methods outperformed the highest single data-type network and the network trained on hand-crafted features. Grad-CAM was shown to more accurately highlight areas of disease.

Conclusions

Ensemble learning increases the predictive accuracy of CNNs for classifying referable DR on OCTA datasets.

Translational Relevance

Because the diagnostic accuracy of OCTA images is shown to be greater than the manually extracted features currently used in the literature, the proposed methods may be beneficial toward developing clinically valuable solutions for DR diagnoses.

Detection and visualization of abnormality in chest radiographs using modality-specific convolutional neural network ensembles

Convolutional neural networks (CNNs) trained on natural images are extremely successful in image classification and localization due to superior automated feature extraction capability. In extending their use to biomedical recognition tasks, it is important to note that visual features of medical images tend to be uniquely different than natural images. There are advantages offered through training these networks on large scale medical common modality image collections pertaining to the recognition task. Further, improved generalization in transferring knowledge across similar tasks is possible when the models are trained to learn modality-specific features and then suitably repurposed for the target task. In this study, we propose modality-specific ensemble learning toward improving abnormality detection in chest X-rays (CXRs). CNN models are trained on a large-scale CXR collection to learn modality-specific features and then repurposed for detecting and localizing abnormalities. Model predictions are combined using different ensemble strategies toward reducing prediction variance and sensitivity to the training data while improving overall performance and generalization. Class-selective relevance mapping (CRM) is used to visualize the learned behavior of the individual models and their ensembles. It localizes discriminative regions of interest (ROIs) showing abnormal regions and offers an improved explanation of model predictions. It was observed that the model ensembles demonstrate superior localization performance in terms of Intersection of Union (IoU) and mean Average Precision (mAP) metrics than any individual constituent model.

A systematic review of the diagnostic accuracy of artificial intelligence-based computer programs to analyze chest x-rays for pulmonary tuberculosis

We undertook a systematic review of the diagnostic accuracy of artificial intelligence-based software for identification of radiologic abnormalities (computer-aided detection, or CAD) compatible with pulmonary tuberculosis on chest x-rays (CXRs). We searched four databases for articles published between January 2005-February 2019. We summarized data on CAD type, study design, and diagnostic accuracy. We assessed risk of bias with QUADAS-2. We included 53 of the 4712 articles reviewed: 40 focused on CAD design methods (“Development” studies) and 13 focused on evaluation of CAD (“Clinical” studies). Meta-analyses were not performed due to methodological differences. Development studies were more likely to use CXR databases with greater potential for bias as compared to Clinical studies. Areas under the receiver operating characteristic curve (median AUC [IQR]) were significantly higher: in Development studies AUC: 0.88 [0.82–0.90]) versus Clinical studies (0.75 [0.66–0.87]; p-value 0.004); and with deep-learning (0.91 [0.88–0.99]) versus machine-learning (0.82 [0.75–0.89]; p = 0.001). We conclude that CAD programs are promising, but the majority of work thus far has been on development rather than clinical evaluation. We provide concrete suggestions on what study design elements should be improved.

Performance evaluation of deep neural ensembles toward malaria parasite detection in thin-blood smear images

Background

Malaria is a life-threatening disease caused by Plasmodium parasites that infect the red blood cells (RBCs). Manual identification and counting of parasitized cells in microscopic thick/thin-film blood examination remains the common, but burdensome method for disease diagnosis. Its diagnostic accuracy is adversely impacted by inter/intra-observer variability, particularly in large-scale screening under resource-constrained settings.

Introduction

State-of-the-art computer-aided diagnostic tools based on data-driven deep learning algorithms like convolutional neural network (CNN) has become the architecture of choice for image recognition tasks. However, CNNs suffer from high variance and may overfit due to their sensitivity to training data fluctuations.

Objective

The primary aim of this study is to reduce model variance, improve robustness and generalization through constructing model ensembles toward detecting parasitized cells in thin-blood smear images.

Methods

We evaluate the performance of custom and pretrained CNNs and construct an optimal model ensemble toward the challenge of classifying parasitized and normal cells in thin-blood smear images. Cross-validation studies are performed at the patient level to ensure preventing data leakage into the validation and reduce generalization errors. The models are evaluated in terms of the following performance metrics: (a) Accuracy; (b) Area under the receiver operating characteristic (ROC) curve (AUC); (c) Mean squared error (MSE); (d) Precision; (e) F-score; and (f) Matthews Correlation Coefficient (MCC).

Results

It is observed that the ensemble model constructed with VGG-19 and SqueezeNet outperformed the state-of-the-art in several performance metrics toward classifying the parasitized and uninfected cells to aid in improved disease screening.

Conclusions

Ensemble learning reduces the model variance by optimally combining the predictions of multiple models and decreases the sensitivity to the specifics of training data and selection of training algorithms. The performance of the model ensemble simulates real-world conditions with reduced variance, overfitting and leads to improved generalization.