Automatic Image Interpretation Research Articles

A domain knowledge-based interpretable deep learning system for improving clinical breast ultrasound diagnosis.

Though deep learning has consistently demonstrated advantages in the automatic interpretation of breast ultrasound images, its black-box nature hinders potential interactions with radiologists, posing obstacles for clinical deployment. We proposed a domain knowledge-based interpretable deep learning system for improving breast cancer risk prediction via paired multimodal ultrasound images. The deep learning system was developed on 4320 multimodal breast ultrasound images of 1440 biopsy-confirmed lesions from 1348 prospectively enrolled patients across two hospitals between August 2019 and December 2022. The lesions were allocated to 70% training cohort, 10% validation cohort, and 20% test cohort based on case recruitment date. Here, we show that the interpretable deep learning system can predict breast cancer risk as accurately as experienced radiologists, with an area under the receiver operating characteristic curve of 0.902 (95% confidence interval = 0.882 - 0.921), sensitivity of 75.2%, and specificity of 91.8% on the test cohort. With the aid of the deep learning system, particularly its inherent explainable features, junior radiologists tend to achieve better clinical outcomes, while senior radiologists experience increased confidence levels. Multimodal ultrasound images augmented with domain knowledge-based reasoning cues enable an effective human-machine collaboration at a high level of prediction performance. Such a clinically applicable deep learning system may be incorporated into future breast cancer screening and support assisted or second-read workflows.

RL-TEE: Autonomous Probe Guidance for Transesophageal Echocardiography Based on Attention-Augmented Deep Reinforcement Learning

Ultrasound image acquisition in conventional transesophageal echocardiography (TEE) requires complex manual operation of the probe in the esophagus based on the interpretation of ultrasound images and in-depth knowledge of the cardiac anatomy. In this work, we formulate the TEE probe guidance task as a reinforcement learning (RL) problem, and present the first learning-based solution to 3-DOF control of a TEE probe based on the ultrasound image feedback, named RL-TEE, in order to mimic the visual search and navigation strategies of expert echocardiographers. The probe-tissue interaction in TEE is carefully modeled in our framework by considering both the requirements for navigation towards the standard views and compliance in the esophageal environment. Furthermore, we propose a hybrid deep Q-network model that augments a convolutional neural network backbone with self-attention mechanisms to better capture spatial information in ultrasound images to guide navigation decisions. The presented methods are preliminarily validated in a TEE simulation environment built with data from 25 subjects to acquire four standard views of the heart. Our results show that the proposed method can effectively learn to accurately and compliantly guide the probe movement for TEE standard view acquisition tasks and has a good generalization ability to unseen patient data. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —The motivation of this paper is to realize 3-DOF movement guidance of a TEE probe to acquire the standard views of the heart based on the real-time images, which can be applied to existing robotic control systems or used to assist novice echocardiographers in TEE examination, thereby relieving operator workload and improving ease of use. This paper suggests a novel approach that uses the deep RL technique to achieve automatic interpretation of TEE images and intelligent guidance of the probe movement. The RL framework is designed to take into account both the navigation efficiency and compliance with the esophageal environment for the targeted intracorporeal application. A hybrid deep Q-network model that augments a convolutional neural network with attention mechanisms is designed to better capture spatial information from ultrasound images to predict the probe movement. The effectiveness of the framework is preliminarily validated in extensive experiments in a simulation environment built with real patient data. The proposed method can be applied in clinical use to provide real-time TEE probe guidance for novice echocardiographers, and can be integrated with a robotic system to fully automate the TEE acquisition, thereby relieving the doctors from tedious manual operation to focus on the diagnosis and treatment.

THE ROLE OF ARTIFICIAL INTELLIGENCE (AI) ON MRI BRAIN EXAMINATION WITH CLINICAL ISCHEMIC STROKE

Background: Application of Artificial Intelligence (AI) in radiology is named automatic image interpretation of neuroimaging stroke. It takes a short time to minimize the patient's brain damage. Purpose: Determine the role of AI in ischemic brain stroke MRI examination and find out the advantages and disadvantages of applying AI to ischemic brain stroke MRI examination. Review: It was a descriptive and qualitative study with a literature review approach. The selection of articles used the ScienceDirect, Scopus, ProQuest, PubMed, and Publish or Perish databases. The inclusion criteria included full articles, with the topic of AI on ischemic brain stroke MRI examinations published in the 2017 – 2022 range, articles published by English-language international journals with a classification of Q1 – Q3, and having DOI. Seven relevant pieces of article were obtained, then descriptive analysis was carried out by comparing and presenting the articles descriptively in tabular form. Result: The role of AI in MRI brain examination with clinical ischemic stroke, namely its role in automatic lesion segmentation, Time Since Stroke (TSS) classification, and infarct volume prediction. The advantages of AI included short image processing times and accurate results. The disadvantages of AI tended to decrease performance in small lesions, a large number of patients, limited data, and false positive results. The value of the Dice Score Coefficient (DSC) (0.53 – 0.86) was already high even though it had not reached 1 because it depended on the strength of the data used. Conclusion: The role of AI in MRI imaging of ischemic brain stroke helps in the diagnosis and prognosis of ischemic stroke patients. AI in stroke neuroimaging has advantages in time effectiveness and disadvantages in data limitations.

End-to-end deep learning model for underground utilities localization using GPR

Underground utilities (UUs) are key infrastructures in urban life operations. The localization of UUs is vital to governments and residents in terms of asset management, utility planning, and construction safety. UUs localization has been investigated extensively via the automatic interpretation of ground-penetrating radar B-scan images. However, conventional image processing methods are time consuming and susceptible to noise. Deep learning-based methods cannot optimize parameters globally because of their box-fitting mode, which requires the separation of a task into region detection and hyperbola fitting problems. Thus, the accuracy and robustness of the localization task are reduced. Hence, an end-to-end deep learning model based on a key point–regression mode is proposed and validated in this study. Experimental results show that the proposed method outperforms the current mainstream models in terms of localization accuracy (97.01%), inference speed (125 fps), and robustness on the same platform (NVDIA RTX 3090 GPU).

Industrial X-ray Image Analysis with Deep Neural Networks Robust to Unexpected Input Data

X-ray inspection is often an essential part of quality control within quality critical manufacturing industries. Within such industries, X-ray image interpretation is resource intensive and typically conducted by humans. An increased level of automatization would be preferable, and recent advances in artificial intelligence (e.g., deep learning) have been proposed as solutions. However, typically, such solutions are overconfident when subjected to new data far from the training data, so-called out-of-distribution (OOD) data; we claim that safe automatic interpretation of industrial X-ray images, as part of quality control of critical products, requires a robust confidence estimation with respect to OOD data. We explored if such a confidence estimation, an OOD detector, can be achieved by explicit modeling of the training data distribution, and the accepted images. For this, we derived an autoencoder model trained unsupervised on a public dataset with X-ray images of metal fusion welds and synthetic data. We explicitly demonstrate the dangers with a conventional supervised learning-based approach and compare it to the OOD detector. We achieve true positive rates of around 90% at false positive rates of around 0.1% on samples similar to the training data and correctly detect some example OOD data.

Research on Void Signal Recognition Algorithm of 3D Ground-Penetrating Radar Based on the Digital Image

The three-dimensional ground-penetrating radar system is an effective method to detect road void disease. Ground penetrating radar image interpretation has the characteristics of multi-solution, long interpretation period, and high professional requirements of processors. In recent years, researchers have put forward solutions for automatic interpretation of ground-penetrating radar images, including automatic detection algorithm for subgrade diseases based on support vector machines, etc., but there are still some shortcomings such as training models with a large amount of data or setting parameters. In this article, a three-dimensional ground-penetrating radar void signal recognition algorithm based on the digital image is proposed, and the algorithm uses digital images to characterize radar signals. With the help of digital image processing methods, the images are processed by binarization, corrosion, expansion, connected area inspection, fine length index inspection, and three-dimensional matching inspection, so as to identify and determine the void signals and extract the void area volume index. The algorithm has been verified by laboratory tests and engineering projects, and the results show that the void identification algorithm can accurately identify the void area position; the error level between the measured values and the measured values of length, width, buried depth, and area is between 2.2 and 17.3%, and the error is generally within the engineering acceptance range. The volume index calculated by the algorithm has a certain engineering application value; compared with the support vector machine, the regressive convolution neural network, and other recognition methods, it has the advantage of not needing a large amount of data to train or modify parameters.

P156 A multicentric study on the development and application of a deep learning algorithm for automatic detection of ulcers and erosions in the novel PillCam™ Crohn’s capsule

Abstract Background Crohn’s disease has the potential to involve all segments of the gastrointestinal tract, requiring frequent and judicious clinical and endoscopic surveillance. Capsule endoscopy (CE) has revolutionized the management of these patients, playing a role in diagnosis, classification, and assessment of response to treatment. In 2017, PillCam™ Crohn’s Capsule (PCC) was introduced to provide a panenteric assessment to this panenteric disease. However, reviewing each PCC exam is time-consuming and prone to errors as it implies examining thousands of images while significant lesions may only be represented in a small number of frames. Recently, the development of artificial intelligence (AI) algorithms for automatic interpretation of endoscopic images has generated a large interest. The performance of these tools, particularly convolutional neural networks have shown potential to overcome some of the main drawbacks of CE. This multicentric study aims to assess the performance of a deep learning algorithm for the automatic detection of small intestinal and colonic ulcers and erosions in PCC images. Methods We included a total of 59 PCC exams from two centres (ManopH Gastroenterology Clinic and Centro Hospitalar Universitário de São João) performed between 2017 and 2021. A total of 78415 frames were extracted. This pool of images was constituted by 14124 images depicting ulcers and erosions and 64291 with normal enteric and colonic mucosa. For the automatic identification of these findings, this image pool was split into training and validation datasets in a patient-based manner. Therefore, images from the same patient were restricted to a single dataset. A CNN model with transfer learning was developed using tensorflow and keras tools. Results The neural architecture of our CNN was optimized for the detection of enteric and colonic ulcers and erosions. An example of the output provided by the CNN is shown in Figure 1. The deep learning model had a sensitivity of 86.7%, a specificity of 98.6%, a positive predictive value of 92.3% and a negative predictive value of 97.4%. The overall accuracy of our network was 96.6% (Figure 2). At a threshold value of 50%, our CNN had an area under the curve of 0.98 for the detection of ulcers and erosions (Figure 3). Conclusion In this multicentric study, our group has developed a CNN and assessed its performance in a patient-based design, which provides significant robustness to our results. Patient-split analyses are the first step towards real-life application of AI solutions which, allied to the high-performance results of our model, may significantly improve the diagnostic yield of this panendoscopic tool.

Non-invasive diagnosis of deep vein thrombosis from ultrasound imaging with machine learning

Deep vein thrombosis (DVT) is a blood clot most commonly found in the leg, which can lead to fatal pulmonary embolism (PE). Compression ultrasound of the legs is the diagnostic gold standard, leading to a definitive diagnosis. However, many patients with possible symptoms are not found to have a DVT, resulting in long referral waiting times for patients and a large clinical burden for specialists. Thus, diagnosis at the point of care by non-specialists is desired. We collect images in a pre-clinical study and investigate a deep learning approach for the automatic interpretation of compression ultrasound images. Our method provides guidance for free-hand ultrasound and aids non-specialists in detecting DVT. We train a deep learning algorithm on ultrasound videos from 255 volunteers and evaluate on a sample size of 53 prospectively enrolled patients from an NHS DVT diagnostic clinic and 30 prospectively enrolled patients from a German DVT clinic. Algorithmic DVT diagnosis performance results in a sensitivity within a 95% CI range of (0.82, 0.94), specificity of (0.70, 0.82), a positive predictive value of (0.65, 0.89), and a negative predictive value of (0.99, 1.00) when compared to the clinical gold standard. To assess the potential benefits of this technology in healthcare we evaluate the entire clinical DVT decision algorithm and provide cost analysis when integrating our approach into diagnostic pathways for DVT. Our approach is estimated to generate a positive net monetary benefit at costs up to £72 to £175 per software-supported examination, assuming a willingness to pay of £20,000/QALY.

POS0256 AUTOMATED CAPILLARY DETECTION AND IMAGE ANALYSIS SOFTWARE IN CAPILLAROSCOPY: CAPILLARY.IO

Background: A nailfold capillaroscopy procedure is a non-invasive, low-cost, and well-established examination that can be used to diagnose several rheumatic autoimmune diseases and support the necessary follow-up of patients. There are two main elements to nailfold capillaroscopy: image acquisition and image interpretation. Both of these can present challenges: we need to ensure that the best possible images are captured, and we need to define the enlarged capillaries, capillary loss and pericapillary hemorrhages objectively. We introduce Capillary.io, an automatic image reading system able to recognize capillaries in images obtained with any microscope, generate automatic measurements of each capillary and take advantage of this information to report capillary morphology. Together it allows a comprehensive analysis that is capable of producing detailed reports for each patient. Objectives: The primary outcome was general sensitivity and specificity, using images assessed by expert capillaroscopists as the gold standard. Methods: 6500 images previously analyzed by capillaroscopists from GREC were compared with Capillary.io. Capillary morphology (enlarged capillaries, tortuosities, ramifications, megacapillaries, hemorrhages) of each of the capillaries contained in each of the images was analyzed manually by at least one expert capillaroscopist. Subsequently, the automatic image interpretation system was used to fully automatically analyze each of the capillaries contained in each image and the results obtained were compared. Results: Overall, a total of 78.347 capillaries were compared, of which 47.734 were normal capillaries, 21.991 enlarged capillaries, 2672 megacapillaries, 8512 tortuosities, 1322 ramifications and 5149 hemorrhages. Capillary.io was able to detect 38.101 normal capillaries, 19.126 enlarged capillaries, 2389 megacapillaries, 5698 tortuosities, 718 ramifications and 3706 hemorraghes. Capillary.io presented a sensitivity (S) of 79.82% and a specificity (E) of 82% in the recognition of normal capillaries. The automatized system was able to recognize enlarged capillaries with a sensitivity of 86.97% and a specificity of 81.38%. Megacapillaries were detected with 89.41% sensitivity and 78.75% specificity. Similarly, the system was able to detect tortuosities (S 66.94%; E 67.71%), ramifications (S 54.34%; E 58.61%) and hemorrhages (S 71.36; E 73.97%). Conclusion: Capillary.io is a simple, easy-learning web-based system to get interpretation of nailfold capillaroscopic images. It may be a very useful tool to standardize the interpretation of capillaroscopic pictures and could provide great research in that field.

Automatic medical image interpretation: State of the art and future directions

Automatic Natural language interpretation of medical images is an emerging field of Artificial Intelligence (AI). The task combines two fields of AI; computer vision and natural language processing. This is a challenging task that goes beyond object detection, segmentation, and classification because it also requires the understanding of the relationship between different objects of an image and the actions performed by these objects as visual representations. Image interpretation is helpful in many tasks like helping visually impaired persons, information retrieval, early childhood learning, producing human like natural interaction between robots, and many more applications. Recently this work fascinated researchers to use the same approach by using more complex biomedical images. It has been applied from generating single sentence captions to multi sentence paragraph descriptions. Medical image captioning can assist and speed up the diagnosis process of medical professionals and generated report can be used for many further tasks. This is a comprehensive review of recent years’ research of medical image captioning published in different international conferences and journals. Their common parameters are extracted to compare their methods, performance, strengths, limitations, and our recommendations are discussed. Further publicly available datasets and evaluation measures used for deep-learning based captioning of medical images are also discussed.

Non-Invasive Diagnosis of Deep Vein Thrombosis From Ultrasound With Machine Learning

Background: Deep Vein Thrombosis (DVT) is a blood clot most found in the leg, which can lead to fatal pulmonary embolism (PE). The diagnostic standard for DVT is a compression ultrasound examination of the legs. Many patients with possible symptoms are not found to have a DVT, resulting in long waiting times for patients and a large clinical burden for specialists. Machine learning has shown great potential for the automatic interpretation of images but has not been evaluated at the front line of care yet. We collect images in a pre-clinical study and investigate a machine-learning software for the interpretation of compression ultrasound images to aid non-specialists in detecting DVT. Methods: Selected anatomical areas are commonly scanned to examine for DVT (common femoral vein, superficial femoral vein, popliteal vein). A machine learning model automatically identifies the veins and arteries and predicts which of the areas, if any, is present in the ultrasound images. Compressions are algorithmically instructed, and compressibility is automatically evaluated. We tested on ultrasound videos scanned by professionals, where each frame has been manually labelled with the correct anatomical location and vessel compression status to quantify method performance. A cost-effectiveness analysis for clinical pathway integration is also provided. Findings: Evaluation was performed on a sample size of 51 prospectively enrolled patients from an NHS DVT diagnostic clinic. 32 DVT-positive patients and 19 DVT-negative patients were included. Algorithmic DVT diagnosis results in a sensitivity of 93.8% and a specificity of 84.2%, a positive predictive value of 90.9%, and a negative predictive value of 88.9% compared to clinical gold standard. Integrating our approach into a clinical DVT pathway is estimated to be cost effective at up to $150 per software examination, assuming a willingness to pay $26 000/QALY. Interpretation: Machine learning methods have good diagnostic accuracy and have the potential to be cost-effective for identifying DVT in patients. Future work will evaluate if front-line-of-care workers can accurately diagnose DVT with our approach. Funding: ThinkSono Ldt. (ThinkSono Ldt funded the development of the method. Data was collected at Oxford Haemophilia & Thrombosis Centre independently. ThinkSono Ldt provided ultrasound data acquisition devices to Oxford Haemophilia & Thrombosis Centre for this study.) Declaration of Interests: B.K., M.H. and N.C. are scientific advisors for ThinkSono Ldt B.K. is also advisor for Ultromics Ldt And Cydar medical Ldt J.O. acts as a consultant for ThinkSono Ldt. A.M. was an employee of ThinkSono Ldt until September 2020. F.AN., S.M. are employees of ThinkSono Ldt, M.D.S. and A.C.R. acted as contractor for ThinkSono Ldt. All authors had full access to all data during this study and accept responsibility to submit for publication. B.K., A.M., F.AN., S.M. are joint inventors on a patent held by ThinkSono Ldt. The views expressed are those of the author(s) and not necessarily those of ThinkSono, the NHS, the NIHR or the Department of Health. Ethics Approval Statement: The University of Oxford, UK, approved the study (Ethics: 18/SC/0220, IRAS 234007). All participants provided written informed consent.

CS2-Net: Deep learning segmentation of curvilinear structures in medical imaging.

Automated detection of curvilinear structures, e.g., blood vessels or nerve fibres, from medical and biomedical images is a crucial early step in automatic image interpretation associated to the management of many diseases. Precise measurement of the morphological changes of these curvilinear organ structures informs clinicians for understanding the mechanism, diagnosis, and treatment of e.g. cardiovascular, kidney, eye, lung, and neurological conditions. In this work, we propose a generic and unified convolution neural network for the segmentation of curvilinear structures and illustrate in several 2D/3D medical imaging modalities. We introduce a new curvilinear structure segmentation network (CS2-Net), which includes a self-attention mechanism in the encoder and decoder to learn rich hierarchical representations of curvilinear structures. Two types of attention modules - spatial attention and channel attention - are utilized to enhance the inter-class discrimination and intra-class responsiveness, to further integrate local features with their global dependencies and normalization, adaptively. Furthermore, to facilitate the segmentation of curvilinear structures in medical images, we employ a 1×3 and a 3×1 convolutional kernel to capture boundary features. Besides, we extend the 2D attention mechanism to 3D to enhance the network's ability to aggregate depth information across different layers/slices. The proposed curvilinear structure segmentation network is thoroughly validated using both 2D and 3D images across six different imaging modalities. Experimental results across nine datasets show the proposed method generally outperforms other state-of-the-art algorithms in various metrics.

Towards Detecting Building Facades with Graffiti Artwork Based on Street View Images

As a recognized type of art, graffiti is a cultural asset and an important aspect of a city’s aesthetics. As such, graffiti is associated with social and commercial vibrancy and is known to attract tourists. However, positional uncertainty and incompleteness are current issues of open geo-datasets containing graffiti data. In this paper, we present an approach towards detecting building facades with graffiti artwork based on the automatic interpretation of images from Google Street View (GSV). It starts with the identification of geo-tagged photos of graffiti artwork posted on the photo sharing media Flickr. GSV images are then extracted from the surroundings of these photos and interpreted by a customized, i.e., transfer learned, convolutional neural network. The compass heading of the GSV images classified as containing graffiti artwork and the possible positions of their acquisition are considered for scoring building facades according to their potential of containing the artwork observable in the GSV images. More than 36,000 GSV images and 5000 facades from buildings represented in OpenStreetMap were processed and evaluated. Precision and recall rates were computed for different facade score thresholds. False-positive errors are caused mostly by advertisements and scribblings on the building facades as well as by movable objects containing graffiti artwork and obstructing the facades. However, considering higher scores as threshold for detecting facades containing graffiti leads to the perfect precision rate. Our approach can be applied for identifying previously unmapped graffiti artwork and for assisting map contributors interested in the topic. Furthermore, researchers interested on the spatial correlations between graffiti artwork and socio-economic factors can profit from our open-access code and results.

Design of Probabilistic Patch Based Glaucoma Detection using CDR by Optic-Disc and Optic-Cup Segmentation

Denoising and Texture features (DTF) within eye images are energetically tracked for more precise and efficient classifications of different diseases in glaucoma images which are captured by very sensitive camera lens. Due to improper lens adjustment and noise usually called speckle and Gaussian noises are the major unwanted component pixels present in the image while capturing and these makes difficult for further accurate diagnosis or automatic image interpretation. To address these issues, this paper develops and demonstrates a new approach for removal of speckle and Gaussian noises using Probabilistic Patch Based (PPB) filter, which is extended version of Non Local means algorithm. The PPB filter is depends on the distribution noise model and maximum weighted likelihood estimation and these weights are training iteratively based on the both noisy patches and patches extracted similarities from the previous estimations as part the this work. In the second part, the features are extracted using Lifting Scheme DWT in terms of low and high frequencies into sub-bands and the required low frequency components are used for further analysis. Cup-to-Disc components are segmented using basic level set function and based on the values of cup and disc values, the glaucoma is affected or not is classified. Comparative analysis has made between proposed results and previously available results and it is found that the present results shows 29% is improvement in accuracy and 15% improvement in accurate identification of glaucoma.

Deep learning-based method for reconstructing three-dimensional building cadastre models from aerial images

Nowadays, many of the world’s large cities are faced with the issue of land scarcity for construction due to the increasing growth of urbanization, as well as the economic downturn for exploiting lands and properties, and city officials have come up with the idea of optimal management of real estate in order to cope with these problems. The purpose of our study is to reconstruct three-dimensional (3-D) building cadastre models (3DBCMs) with an approach to improve the state of land administration in Tehran metropolis. Our study is being implemented and evaluated in three stages. The first stage involves collecting aerial images. The interior and exterior orientation parameters are preprepared in this step. The second stage involves automatic interpretation and extraction of buildings from aerial images by providing a method of interpretation called fully automatic interpretation with deep learning (FAIDL). The third stage involves 3-D building modeling and evaluating the effect of FAIDL method on the automatic interpretation of images. The results showed that the 3-D models of building have a better geometric accuracy compare to 60 cm, which are produced with the proposed algorithm.

A Machine Learning Based Approach for Automatic Rebar Detection and Quantification of Deterioration in Concrete Bridge Deck Ground Penetrating Radar B-scan Images

Ground penetrating radar (GPR) is a non-destructive method (NDT) for subsurface object identification. Interpretation of GPR data is often done manually by an engineer, which is a time-intensive task and requires moderate to significant level of training. The authors proposed a novel machine learning based processing for automatic interpretation and quantification of concrete bridge deck GPR B-scan images. The proposed method is based on combination of image processing, machine learning (ML) data classification, data filtering, and spatial pattern analysis for quantification of deterioration in concrete bridge decks. For the first time, the authors introduced a dataset of 4,000 B-scan images cropped from real bridge deck GPR field data, named DECKGPRH1.0. The proposed method is tested on bridge deck GPR data collected from three bridges with different NBI (National Bridge Inventory) ratings. The results presented indicate that by implementing a ML based classifier and a fine tuned filter, the proposed approach provides a robust solution for automatic quantification GPR field data.

Automatic diagnostics of tuberculosis using convolutional neural networks analysis of MODS digital images.

Tuberculosis is an infectious disease that causes ill health and death in millions of people each year worldwide. Timely diagnosis and treatment is key to full patient recovery. The Microscopic Observed Drug Susceptibility (MODS) is a test to diagnose TB infection and drug susceptibility directly from a sputum sample in 7–10 days with a low cost and high sensitivity and specificity, based on the visual recognition of specific growth cording patterns of M. Tuberculosis in a broth culture. Despite its advantages, MODS is still limited in remote, low resource settings, because it requires permanent and trained technical staff for the image-based diagnostics. Hence, it is important to develop alternative solutions, based on reliable automated analysis and interpretation of MODS cultures. In this study, we trained and evaluated a convolutional neural network (CNN) for automatic interpretation of MODS cultures digital images. The CNN was trained on a dataset of 12,510 MODS positive and negative images obtained from three different laboratories, where it achieved 96.63 +/- 0.35% accuracy, and a sensitivity and specificity ranging from 91% to 99%, when validated across held-out laboratory datasets. The model's learned features resemble visual cues used by expert diagnosticians to interpret MODS cultures, suggesting that our model may have the ability to generalize and scale. It performed robustly when validated across held-out laboratory datasets and can be improved upon with data from new laboratories. This CNN can assist laboratory personnel, in low resource settings, and is a step towards facilitating automated diagnostics access to critical areas in developing countries.

Monitoring of beach litter by automatic interpretation of unmanned aerial vehicle images using the segmentation threshold method

This study was aimed at monitoring beach litter using an unmanned aerial vehicle (UAV) in the coastal city of Fuzhou, China. The data analysis shows that the optical images obtained by digital cameras on the UAV can help to identify and monitor beach litter using remote sensing and GIS technologies. The threshold method can effectively segment the UAV image in the beach area. It is useful for quickly monitoring the distribution of beach litter in the area of interest, and hence it can help to provide effective technical support for the investigation and assessment of coastal beach litter.

Characterization of coronary artery pathological formations from OCT imaging using deep learning.

Coronary artery disease is the number one health hazard leading to the pathological formations in coronary artery tissues. In severe cases, they can lead to myocardial infarction and sudden death. Optical Coherence Tomography (OCT) is an interferometric imaging modality, which has been recently used in cardiology to characterize coronary artery tissues providing high resolution ranging from 10 to 20 µm. In this study, we investigate different deep learning models for robust tissue characterization to learn the various intracoronary pathological formations caused by Kawasaki disease (KD) from OCT imaging. The experiments are performed on 33 retrospective cases comprising of pullbacks of intracoronary cross-sectional images obtained from different pediatric patients with KD. Our approach evaluates deep features computed from three different pre-trained convolutional networks. Then, a majority voting approach is applied to provide the final classification result. The results demonstrate high values of accuracy, sensitivity, and specificity for each tissue (up to 0.99 ± 0.01). Hence, deep learning models and especially, majority voting method are robust for automatic interpretation of the OCT images.

Geospatial Object Detection in Remote Sensing Imagery Based on Multiscale Single-Shot Detector with Activated Semantics

Geospatial object detection from high spatial resolution (HSR) remote sensing imagery is a heated and challenging problem in the field of automatic image interpretation. Despite convolutional neural networks (CNNs) having facilitated the development in this domain, the computation efficiency under real-time application and the accurate positioning on relatively small objects in HSR images are two noticeable obstacles which have largely restricted the performance of detection methods. To tackle the above issues, we first introduce semantic segmentation-aware CNN features to activate the detection feature maps from the lowest level layer. In conjunction with this segmentation branch, another module which consists of several global activation blocks is proposed to enrich the semantic information of feature maps from higher level layers. Then, these two parts are integrated and deployed into the original single shot detection framework. Finally, we use the modified multi-scale feature maps with enriched semantics and multi-task training strategy to achieve end-to-end detection with high efficiency. Extensive experiments and comprehensive evaluations on a publicly available 10-class object detection dataset have demonstrated the superiority of the presented method.