We demonstrate use of iteratively pruned deep learning model ensembles for detecting pulmonary manifestations of COVID-19 with chest X-rays. This disease is caused by the novel Severe Acute ...Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) virus, also known as the novel Coronavirus (2019-nCoV). A custom convolutional neural network and a selection of ImageNet pretrained models are trained and evaluated at patient-level on publicly available CXR collections to learn modality-specific feature representations. The learned knowledge is transferred and fine-tuned to improve performance and generalization in the related task of classifying CXRs as normal, showing bacterial pneumonia, or COVID-19-viral abnormalities. The best performing models are iteratively pruned to reduce complexity and improve memory efficiency. The predictions of the best-performing pruned models are combined through different ensemble strategies to improve classification performance. Empirical evaluations demonstrate that the weighted average of the best-performing pruned models significantly improves performance resulting in an accuracy of 99.01% and area under the curve of 0.9972 in detecting COVID-19 findings on CXRs. The combined use of modality-specific knowledge transfer, iterative model pruning, and ensemble learning resulted in improved predictions. We expect that this model can be quickly adopted for COVID-19 screening using chest radiographs.
Data-driven deep learning (DL) methods using convolutional neural networks (CNNs) demonstrate promising performance in natural image computer vision tasks. However, their use in medical computer ...vision tasks faces several limitations, viz., (i) adapting to visual characteristics that are unlike natural images; (ii) modeling random noise during training due to stochastic optimization and backpropagation-based learning strategy; (iii) challenges in explaining DL black-box behavior to support clinical decision-making; and (iv) inter-reader variability in the ground truth (GT) annotations affecting learning and evaluation. This study proposes a systematic approach to address these limitations through application to the pandemic-caused need for Coronavirus disease 2019 (COVID-19) detection using chest X-rays (CXRs). Specifically, our contribution highlights significant benefits obtained through (i) pretraining specific to CXRs in transferring and fine-tuning the learned knowledge toward improving COVID-19 detection performance; (ii) using ensembles of the fine-tuned models to further improve performance over individual constituent models; (iii) performing statistical analyses at various learning stages for validating results; (iv) interpreting learned individual and ensemble model behavior through class-selective relevance mapping (CRM)-based region of interest (ROI) localization; and, (v) analyzing inter-reader variability and ensemble localization performance using Simultaneous Truth and Performance Level Estimation (STAPLE) methods. We find that ensemble approaches markedly improved classification and localization performance, and that inter-reader variability and performance level assessment helps guide algorithm design and parameter optimization. To the best of our knowledge, this is the first study to construct ensembles, perform ensemble-based disease ROI localization, and analyze inter-reader variability and algorithm performance for COVID-19 detection in CXRs.
Full text
Available for:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Artificial intelligence (AI) continues to garner substantial interest in medical imaging. The potential applications are vast and include the entirety of the medical imaging life cycle from image ...creation to diagnosis to outcome prediction. The chief obstacles to development and clinical implementation of AI algorithms include availability of sufficiently large, curated, and representative training data that includes expert labeling (eg, annotations). Current supervised AI methods require a curation process for data to optimally train, validate, and test algorithms. Currently, most research groups and industry have limited data access based on small sample sizes from small geographic areas. In addition, the preparation of data is a costly and time-intensive process, the results of which are algorithms with limited utility and poor generalization. In this article, the authors describe fundamental steps for preparing medical imaging data in AI algorithm development, explain current limitations to data curation, and explore new approaches to address the problem of data availability.
To evaluate the performance of a prototype photon-counting detector (PCD) computed tomography (CT) system for abdominal CT in humans and to compare the results with a conventional energy-integrating ...detector (EID).
The study was HIPAA-compliant and institutional review board-approved with informed consent. Fifteen asymptomatic volunteers (seven men; mean age, 58.2 years ± 9.8 standard deviation) were prospectively enrolled between September 2 and November 13, 2015. Radiation dose-matched delayed contrast agent-enhanced spiral and axial abdominal EID and PCD scans were acquired. Spiral images were scored for image quality (Wilcoxon signed-rank test) in five regions of interest by three radiologists blinded to the detector system, and the axial scans were used to assess Hounsfield unit accuracy in seven regions of interest (paired t test). Intraclass correlation coefficient (ICC) was used to assess reproducibility. PCD images were also used to calculate iodine concentration maps. Spatial resolution, noise-power spectrum, and Hounsfield unit accuracy of the systems were estimated by using a CT phantom.
In both systems, scores were similar for image quality (median score, 4; P = .19), noise (median score, 3; P = .30), and artifact (median score, 1; P = .17), with good interrater agreement (image quality, noise, and artifact ICC: 0.84, 0.88, and 0.74, respectively). Hounsfield unit values, spatial resolution, and noise-power spectrum were also similar with the exception of mean Hounsfield unit value in the spinal canal, which was lower in the PCD than the EID images because of beam hardening (20 HU vs 36.5 HU; P < .001). Contrast-to-noise ratio of enhanced kidney tissue was improved with PCD iodine mapping compared with EID (5.2 ± 1.3 vs 4.0 ± 1.3; P < .001).
The performance of PCD showed no statistically significant difference compared with EID when the abdomen was evaluated in a conventional scan mode. PCD provides spectral information, which may be used for material decomposition.
The p110δ subunit of phosphatidylinositol-3-OH kinase (PI(3)K) is selectively expressed in leukocytes and is critical for lymphocyte biology. Here we report fourteen patients from seven families who ...were heterozygous for three different germline, gain-of-function mutations in PIK3CD (which encodes p110δ). These patients presented with sinopulmonary infections, lymphadenopathy, nodular lymphoid hyperplasia and viremia due to cytomegalovirus (CMV) and/or Epstein-Barr virus (EBV). Strikingly, they had a substantial deficiency in naive T cells but an over-representation of senescent effector T cells. In vitro, T cells from patients exhibited increased phosphorylation of the kinase Akt and hyperactivation of the metabolic checkpoint kinase mTOR, enhanced glucose uptake and terminal effector differentiation. Notably, treatment with rapamycin to inhibit mTOR activity in vivo partially restored the abundance of naive T cells, largely 'rescued' the in vitro T cell defects and improved the clinical course.
Full text
Available for:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
The computer-based process of identifying the boundaries of lung from surrounding thoracic tissue on computed tomographic (CT) images, which is called segmentation, is a vital first step in ...radiologic pulmonary image analysis. Many algorithms and software platforms provide image segmentation routines for quantification of lung abnormalities; however, nearly all of the current image segmentation approaches apply well only if the lungs exhibit minimal or no pathologic conditions. When moderate to high amounts of disease or abnormalities with a challenging shape or appearance exist in the lungs, computer-aided detection systems may be highly likely to fail to depict those abnormal regions because of inaccurate segmentation methods. In particular, abnormalities such as pleural effusions, consolidations, and masses often cause inaccurate lung segmentation, which greatly limits the use of image processing methods in clinical and research contexts. In this review, a critical summary of the current methods for lung segmentation on CT images is provided, with special emphasis on the accuracy and performance of the methods in cases with abnormalities and cases with exemplary pathologic findings. The currently available segmentation methods can be divided into five major classes: (a) thresholding-based, (b) region-based, (c) shape-based, (d) neighboring anatomy-guided, and (e) machine learning-based methods. The feasibility of each class and its shortcomings are explained and illustrated with the most common lung abnormalities observed on CT images. In an overview, practical applications and evolving technologies combining the presented approaches for the practicing radiologist are detailed.
Research on detecting Tuberculosis (TB) findings on chest radiographs (or Chest X-rays: CXR) using convolutional neural networks (CNNs) has demonstrated superior performance due to the emergence of ...publicly available, large-scale datasets with expert annotations and availability of scalable computational resources. However, these studies use only the frontal CXR projections, i.e., the posterior-anterior (PA), and the anterior-posterior (AP) views for analysis and decision-making. Lateral CXRs which are heretofore not studied help detect clinically suspected pulmonary TB, particularly in children. Further, Vision Transformers (ViTs) with built-in self-attention mechanisms have recently emerged as a viable alternative to the traditional CNNs. Although ViTs demonstrated notable performance in several medical image analysis tasks, potential limitations exist in terms of performance and computational efficiency, between the CNN and ViT models, necessitating a comprehensive analysis to select appropriate models for the problem under study. This study aims to detect TB-consistent findings in lateral CXRs by constructing an ensemble of the CNN and ViT models. Several models are trained on lateral CXR data extracted from two large public collections to transfer modality-specific knowledge and fine-tune them for detecting findings consistent with TB. We observed that the weighted averaging ensemble of the predictions of CNN and ViT models using the optimal weights computed with the Sequential Least-Squares Quadratic Programming method delivered significantly superior performance (MCC: 0.8136, 95% confidence intervals (CI): 0.7394, 0.8878,
< 0.05) compared to the individual models and other ensembles. We also interpreted the decisions of CNN and ViT models using class-selective relevance maps and attention maps, respectively, and combined them to highlight the discriminative image regions contributing to the final output. We observed that (i) the model accuracy is not related to disease region of interest (ROI) localization and (ii) the bitwise-AND of the heatmaps of the top-2-performing models delivered significantly superior ROI localization performance in terms of mean average precision mAP@(0.1 0.6) = 0.1820, 95% CI: 0.0771,0.2869,
< 0.05, compared to other individual models and ensembles. The code is available at https://github.com/sivaramakrishnan-rajaraman/Ensemble-of-CNN-and-ViT-for-TB-detection-in-lateral-CXR.
There is consistent demand for clinical exposure from students interested in radiology; however, the COVID-19 pandemic resulted in fewer available options and limited student access to radiology ...departments. Additionally, there is increased demand for radiologists to manage more complex quantification in reports on patients enrolled in clinical trials. We present an online educational curriculum that addresses both of these gaps by virtually immersing students (radiology preprocessors, or RPs) into radiologists’ workflows where they identify and measure target lesions in advance of radiologists, streamlining report quantification. RPs switched to remote work at the beginning of the COVID-19 pandemic in our
National Institutes of Health (NIH)
. We accommodated them by transitioning our curriculum on cross-sectional anatomy and advanced PACS tools to a publicly available online curriculum. We describe collaborations between multiple academic research centers and industry through contributions of academic content to this curriculum. Further, we describe how we objectively assess educational effectiveness with cross-sectional anatomical quizzes and decreasing RP miss rates as they gain experience. Our RP curriculum generated significant interest evidenced by a dozen academic and research institutes providing online presentations including radiology modality basics and quantification in clinical trials. We report a decrease in RP miss rate percentage, including one virtual RP over a period of 1 year. Results reflect training effectiveness through decreased discrepancies with radiologist reports and improved tumor identification over time. We present our RP curriculum and multicenter experience as a pilot experience in a clinical trial research setting. Students are able to obtain useful clinical radiology experience in a virtual learning environment by immersing themselves into a clinical radiologist’s workflow. At the same time, they help radiologists improve patient care with more valuable quantitative reports, previously shown to improve radiologist efficiency. Students identify and measure lesions in clinical trials before radiologists, and then review their reports for self-evaluation based on included measurements from the radiologists. We consider our virtual approach as a supplement to student education while providing a model for how artificial intelligence will improve patient care with more consistent quantification while improving radiologist efficiency.
Full text
Available for:
NUK, SBMB, SBNM, UL, UM, UPUK, VSZLJ
Purpose To investigate whether photon-counting detector (PCD) technology can improve dose-reduced chest computed tomography (CT) image quality compared with that attained with conventional ...energy-integrating detector (EID) technology in vivo. Materials and Methods This was a HIPAA-compliant institutional review board-approved study, with informed consent from patients. Dose-reduced spiral unenhanced lung EID and PCD CT examinations were performed in 30 asymptomatic volunteers in accordance with manufacturer-recommended guidelines for CT lung cancer screening (120-kVp tube voltage, 20-mAs reference tube current-time product for both detectors). Quantitative analysis of images included measurement of mean attenuation, noise power spectrum (NPS), and lung nodule contrast-to-noise ratio (CNR). Images were qualitatively analyzed by three radiologists blinded to detector type. Reproducibility was assessed with the intraclass correlation coefficient (ICC). McNemar, paired t, and Wilcoxon signed-rank tests were used to compare image quality. Results Thirty study subjects were evaluated (mean age, 55.0 years ± 8.7 standard deviation; 14 men). Of these patients, 10 had a normal body mass index (BMI) (BMI range, 18.5-24.9 kg/m
; group 1), 10 were overweight (BMI range, 25.0-29.9 kg/m
; group 2), and 10 were obese (BMI ≥30.0 kg/m
, group 3). PCD diagnostic quality was higher than EID diagnostic quality (P = .016, P = .016, and P = .013 for readers 1, 2, and 3, respectively), with significantly better NPS and image quality scores for lung, soft tissue, and bone and with fewer beam-hardening artifacts (all P < .001). Image noise was significantly lower for PCD images in all BMI groups (P < .001 for groups 1 and 3, P < .01 for group 2), with higher CNR for lung nodule detection (12.1 ± 1.7 vs 10.0 ± 1.8, P < .001). Inter- and intrareader reproducibility were good (all ICC > 0.800). Conclusion Initial human experience with dose-reduced PCD chest CT demonstrated lower image noise compared with conventional EID CT, with better diagnostic quality and lung nodule CNR.
RSNA, 2017 Online supplemental material is available for this article.
Treatment of pulmonary nontuberculous mycobacteria, especially Mycobacterium abscessus, requires prolonged, multidrug regimens with high toxicity and suboptimal efficacy. Options for refractory ...disease are limited.
We reviewed the efficacy and toxicity of inhaled amikacin in patients with treatment-refractory nontuberculous mycobacterial lung disease.
Records were queried to identify patients who had inhaled amikacin added to failing regimens. Lower airway microbiology, symptoms, and computed tomography scan changes were assessed together with reported toxicity.
The majority (80%) of the 20 patients who met entry criteria were women; all had bronchiectasis, two had cystic fibrosis and one had primary ciliary dyskinesia. At initiation of inhaled amikacin, 15 were culture positive for M. abscessus and 5 for Mycobacterium avium complex and had received a median (range) of 60 (6, 190) months of mycobacterial treatment. Patients were followed for a median of 19 (1, 50) months. Eight (40%) patients had at least one negative culture and 5 (25%) had persistently negative cultures. A decrease in smear quantity was noted in 9 of 20 (45%) and in mycobacterial culture growth for 10 of 19 (53%). Symptom scores improved in nine (45%), were unchanged in seven (35%), and worsened in four (20%). Improvement on computed tomography scans was noted in 6 (30%), unchanged in 3 (15%), and worsened in 11 (55%). Seven (35%) stopped amikacin due to: ototoxicity in two (10%), hemoptysis in two (10%), and nephrotoxicity, persistent dysphonia, and vertigo in one each.
In some patients with treatment-refractory pulmonary nontuberculous mycobacterial disease, the addition of inhaled amikacin was associated with microbiologic and/or symptomatic improvement; however, toxicity was common. Prospective evaluation of inhaled amikacin for mycobacterial disease is warranted.