Since the beginning of the COVID-19 pandemic, many different machine learning models have been developed to detect and verify COVID-19 pneumonia based on chest X-ray images. Although promising, ...binary models have only limited implications for medical treatment, whereas the prediction of disease severity suggests more suitable and specific treatment options. In this study, we publish severity scores for the 2358 COVID-19 positive images in the COVIDx8B dataset, creating one of the largest collections of publicly available COVID-19 severity data. Furthermore, we train and evaluate deep learning models on the newly created dataset to provide a first benchmark for the severity classification task. One of the main challenges of this dataset is the skewed class distribution, resulting in undesirable model performance for the most severe cases. We therefore propose and examine different augmentation strategies, specifically targeting majority and minority classes. Our augmentation strategies show significant improvements in precision and recall values for the rare and most severe cases. While the models might not yet fulfill medical requirements, they serve as an appropriate starting point for further research with the proposed dataset to optimize clinical resource allocation and treatment.
In medical imaging, deep learning models can be a critical tool to shorten time-to-diagnosis and support specialized medical staff in clinical decision making. The successful training of deep ...learning models usually requires large amounts of quality data, which are often not available in many medical imaging tasks. In this work we train a deep learning model on university hospital chest X-ray data, containing 1082 images. The data was reviewed, differentiated into 4 causes for pneumonia, and annotated by an expert radiologist. To successfully train a model on this small amount of complex image data, we propose a special knowledge distillation process, which we call Human Knowledge Distillation. This process enables deep learning models to utilize annotated regions in the images during the training process. This form of guidance by a human expert improves model convergence and performance. We evaluate the proposed process on our study data for multiple types of models, all of which show improved results. The best model of this study, called PneuKnowNet, shows an improvement of + 2.3% points in overall accuracy compared to a baseline model and also leads to more meaningful decision regions. Utilizing this implicit data quality-quantity trade-off can be a promising approach for many scarce data domains beyond medical imaging.
The successful integration of neural networks in a clinical setting is still uncommon despite major successes achieved by artificial intelligence in other domains. This is mainly due to the black box ...characteristic of most optimized models and the undetermined generalization ability of the trained architectures. The current work tackles both issues in the radiology domain by focusing on developing an effective and interpretable cardiomegaly detection architecture based on segmentation models. The architecture consists of two distinct neural networks performing the segmentation of both cardiac and thoracic areas of a radiograph. The respective segmentation outputs are subsequently used to estimate the cardiothoracic ratio, and the corresponding radiograph is classified as a case of cardiomegaly based on a given threshold. Due to the scarcity of pixel-level labeled chest radiographs, both segmentation models are optimized in a semi-supervised manner. This results in a significant reduction in the costs of manual annotation. The resulting segmentation outputs significantly improve the interpretability of the architecture's final classification results. The generalization ability of the architecture is assessed in a cross-domain setting. The assessment shows the effectiveness of the semi-supervised optimization of the segmentation models and the robustness of the ensuing classification architecture.
Imaging of the temporal bone and middle ear is challenging for radiologists due to the abundance of distinct anatomical structures and the plethora of possible pathologies. The basis for a precise ...diagnosis is knowledge of the underlying anatomy as well as the clinical presentation and the individual patient's otological status. In this article, we aimed to summarize the most common inflammatory lesions of the temporal bone and middle ear, describe their specific imaging characteristics, and highlight their differential diagnoses. First, we introduce anatomical and imaging fundamentals. Additionally, a point-to-point comparison of the radiological and histological features of the wide spectrum of inflammatory diseases of the temporal bone and middle ear in context with a review of the current literature and current trends is given.
In medical imaging, deep learning models serve as invaluable tools for expediting diagnoses and aiding specialized medical professionals in making clinical decisions. However, effectively training ...deep learning models typically necessitates substantial quantities of high-quality data, a resource often lacking in numerous medical imaging scenarios. One way to overcome this deficiency is to artificially generate such images. Therefore, in this comparative study we train five generative models to artificially increase the amount of available data in such a scenario. This synthetic data approach is evaluated on a a downstream classification task, predicting four causes for pneumonia as well as healthy cases on 1082 chest X-ray images. Quantitative and medical assessments show that a Generative Adversarial Network (GAN)-based approach significantly outperforms more recent diffusion-based approaches on this limited dataset with better image quality and pathological plausibility. We show that better image quality surprisingly does not translate to improved classification performance by evaluating five different classification models and varying the amount of additional training data. Class-specific metrics like precision, recall, and F1-score show a substantial improvement by using synthetic images, emphasizing the data rebalancing effect of less frequent classes. However, overall performance does not improve for most models and configurations, except for a DreamBooth approach which shows a +0.52 improvement in overall accuracy. The large variance of performance impact in this study suggests a careful consideration of utilizing generative models for limited data scenarios, especially with an unexpected negative correlation between image quality and downstream classification improvement.
Pulmonary involvement is common in several infectious and non-infectious diagnostic settings. Imaging findings consistently overlap and are therefore difficult to differentiate by chest-CT. The aim ...of this study was to evaluate the role of CT-textural features(CTTA) for discrimination between atypical viral (respiratory-syncitial-virus(RSV) and herpes-simplex-1-virus (HSV1)), fungal (pneumocystis-jirovecii-pneumonia(PJP)) interstitial pneumonias and alveolar hemorrhage.
By retrospective single-centre analysis we identified 46 consecutive patients (29 m) with RSV(n = 5), HSV1(n = 6), PJP(n = 21) and lung hemorrhage(n = 14) who underwent unenhanced chest CTs in early stages of the disease between 01/2016 and 02/2017. All cases were confirmed by microbiologic direct analysis of bronchial lavage. On chest-CT-scans, the presence of imaging features like ground-glass opacity(GGO), crazy-paving, air-space consolidation, reticulation, bronchial wall thickening and centrilobular nodules were described. A representative large area was chosen in both lungs and used for CTTA-parameters (included heterogeneity, intensity, average, deviation, skewness).
Discriminatory CTTA-features were found between alveolar hemorrhage and PJP consisting of differences in mean heterogeneity(p < 0.015) and uniformity of skewness(p < 0.006). There was no difference between CT-textural features of diffuse alveolar hemorrhage and viral pneumonia or PJP and viral pneumonia. Visual HRCT-assessment yielded great overlap of imaging findings with predominance of GGO for PJP and airspace consolidation for pneumonia/alveolar hemorrhage. Significant correlations between HRCT-based imaging findings and CT-textural features were found for all three disease groups.
CT-textural features showed significant differences in mean heterogeneity and uniformity of skewness. HRCT-based imaging findings correlated with certain CT-textural features showing that the latter have the potential to characterize structural properties of lung parenchyma and related abnormalities.
Purpose
Osteoporosis is prevalent and entails alterations of vertebral bone and marrow. Yet, the spine is also a common site of metastatic spread. Parameters that can be non-invasively measured and ...could capture these alterations are the volumetric bone mineral density (vBMD), proton density fat fraction (PDFF) as an estimate of relative fat content, and failure displacement and load from finite element analysis (FEA) for assessment of bone strength. This study’s purpose was to investigate if osteoporotic and osteoblastic metastatic changes in lumbar vertebrae can be differentiated based on the abovementioned parameters (vBMD, PDFF, and measures from FEA), and how these parameters correlate with each other.
Materials and Methods
Seven patients (3 females, median age: 77.5 years) who received 3-Tesla magnetic resonance imaging (MRI) and multi-detector computed tomography (CT) of the lumbar spine and were diagnosed with either osteoporosis (4 patients) or diffuse osteoblastic metastases (3 patients) were included. Chemical shift encoding-based water-fat MRI (CSE-MRI) was used to extract the PDFF, while vBMD was extracted after automated vertebral body segmentation using CT. Segmentation masks were used for FEA-based failure displacement and failure load calculations. Failure displacement, failure load, and PDFF were compared between patients with osteoporotic vertebrae versus patients with osteoblastic metastases, considering non-fractured vertebrae (L1-L4). Associations between those parameters were assessed using Spearman correlation.
Results
Median vBMD was 59.3 mg/cm
3
in osteoporotic patients. Median PDFF was lower in the metastatic compared to the osteoporotic patients (11.9%
vs
. 43.8%, p=0.032). Median failure displacement and failure load were significantly higher in metastatic compared to osteoporotic patients (0.874 mm
vs
. 0.348 mm, 29,589 N
vs
. 3,095 N, p=0.034 each). A strong correlation was noted between PDFF and failure displacement (rho -0.679, p=0.094). A very strong correlation was noted between PDFF and failure load (rho -0.893, p=0.007).
Conclusion
PDFF as well as failure displacement and load allowed to distinguish osteoporotic from diffuse osteoblastic vertebrae. Our findings further show strong associations between PDFF and failure displacement and load, thus may indicate complimentary pathophysiological associations derived from two non-invasive techniques (CSE-MRI and CT) that inherently measure different properties of vertebral bone and marrow.
Background: Before surgical or transcatheter aortic valve implantation (TAVI), coronary status evaluation is required. The role of combined computed coronary tomography angiography (cCTA) and TAVI ...planning CT in this context is not yet well elucidated. This study assessed whether relevant proximal coronary disease requiring coronary revascularization can be safely detected by combined cCTA and TAVI planning CT, including CT-derived fractional flow reserve (FFR) calculation in patients with severe aortic stenosis. Methods: This study analyzed patients with successful cCTA combined with TAVI planning CT using a 128-slice dual-source scanner. The detection via cCTA of relevant left main stem stenosis (>50%) or proximal coronary artery stenosis (>70%) was compared to invasive coronary angiography (ICA). Results: This study comprised 101 consecutive TAVI patients with a median age of 83 77–86 years, a median STS score of 3.7 2.4–6.1 and 54% of whom had known coronary artery disease. Of 15 patients with relevant coronary stenoses, 14 (93.3%) were detected with cCTA, while false positive results were found in 25 patients. Only in patients with previous percutaneous coronary stent implantation (PCI) were false positive rates (11/29) increased. In the subgroup without previous PCI, an improved classification performance of 87.5%, being mainly due to 11.1% false positive classifications, led to a negative predictive value of 98.5%. Conclusions: Combined cCTA and CT-FFR with TAVI planning CT via state-of-the-art scanners and protocols as a one-stop shop can replace routine ICA in patients prior to TAVI due to its safe detection of relevant coronary artery stenosis, although diagnostic performance of cCTA is only reduced in patients with coronary stents.