Frontiers in radiology最新文献

Artificial Intelligence (AI) methods, particularly Deep Neural Networks (DNNs), have shown great promise in a range of medical imaging tasks. However, the susceptibility of DNNs to producing erroneous outputs under the presence of input noise and variations is of great concern and one of the largest challenges to their adoption in medical settings. Towards addressing this challenge, we explore the robustness of DNNs trained for chest radiograph classification under a range of perturbations reflective of clinical settings. We propose RoMIA, a framework for the creation of Robust Medical Imaging AI models. RoMIA adds three key steps to the model training and deployment flow: (i) Noise-added training, wherein a part of the training data is synthetically transformed to represent common noise sources, (ii) Fine-tuning with input mixing, in which the model is refined with inputs formed by mixing data from the original training set with a small number of images from a different source, and (iii) DCT-based denoising, which removes a fraction of high-frequency components of each image before applying the model to classify it. We applied RoMIA to create six different robust models for classifying chest radiographs using the CheXpert dataset. We evaluated the models on the CheXphoto dataset, which consists of naturally and synthetically perturbed images intended to evaluate robustness. Models produced by RoMIA show 3%-5% improvement in robust accuracy, which corresponds to an average reduction of 22.6% in misclassifications. These results suggest that RoMIA can be a useful step towards enabling the adoption of AI models in medical imaging applications.

Rationale and objectives: We aimed to evaluate the impact of four-dimensional noise reduction filtering using a four-dimensional similarity filter (4D-SF) on radiation dose reduction in dynamic myocardial computed tomography perfusion (CTP).

Materials and methods: Forty-three patients who underwent dynamic myocardial CTP using 320-row computed tomography (CT) were included in the study. The original images were reconstructed using iterative reconstruction (IR). Three different CTP datasets with simulated noise, corresponding to 25%, 50%, and 75% reduction of the original dose (300 mA), were reconstructed using a combination of IR and 4D-SF. The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were assessed, and CT-derived myocardial blood flow (CT-MBF) was quantified. The results were compared between the original and simulated images with radiation dose reduction.

Results: The median SNR (first quartile-third quartile) at the original, 25%-, 50%-, and 75%-dose reduced-simulated images with 4D-SF was 8.3 (6.5-10.2), 16.5 (11.9-21.7), 15.6 (11.0-20.1), and 12.8 (8.8-18.1) and that of CNR was 4.4 (3.2-5.8), 6.7 (4.6-10.3), 6.6 (4.3-10.1), and 5.5 (3.5-9.1), respectively. All the dose-reduced-simulated CTPs with 4D-SF had significantly higher image quality scores in SNR and CNR than the original ones (25%-, 50%-, and 75%-dose reduced vs. original images, p < 0.05, in each). The CT-MBF in 75%-dose reduced-simulated CTP was significantly lower than 25%-, 50%- dose-reduced-simulated, and original CTPs (vs. 75%-dose reduced-simulated images, p < 0.05, in each).

Conclusion: 4D-SF has the potential to reduce the radiation dose associated with dynamic myocardial CTP imaging by half, without impairing the robustness of MBF quantification.

Unlocking the vast potential of deep learning-based computer vision classification systems necessitates large data sets for model training. Natural Language Processing (NLP)-involving automation of dataset labelling-represents a potential avenue to achieve this. However, many aspects of NLP for dataset labelling remain unvalidated. Expert radiologists manually labelled over 5,000 MRI head reports in order to develop a deep learning-based neuroradiology NLP report classifier. Our results demonstrate that binary labels (normal vs. abnormal) showed high rates of accuracy, even when only two MRI sequences (T2-weighted and those based on diffusion weighted imaging) were employed as opposed to all sequences in an examination. Meanwhile, the accuracy of more specific labelling for multiple disease categories was variable and dependent on the category. Finally, resultant model performance was shown to be dependent on the expertise of the original labeller, with worse performance seen with non-expert vs. expert labellers.

Introduction: Osteoporosis (OP) results in weak bone and can ultimately lead to fracture. MRI assessment of bone structure and microarchitecture has been proposed as method to assess bone quality and fracture risk in vivo. Radiomics provides a framework to analyze the textural information of MR images. The purpose of this study was to analyze the radiomic features and its abilityto differentiate between subjects with and without prior fragility fracture.

Methods: MRI acquisition was performed on n = 45 female OP subjects: 15 with fracture history (Fx) and 30 without fracture history (nFx) using a high-resolution 3D Fast Low Angle Shot (FLASH) sequence at 3T. Second and first order radiomic features were calculated in the trabecular region of the proximal femur on T1-weighted MRI signal of a matched dataset. Significance of the feature's predictive ability was measured using Wilcoxon test and Area Under the ROC (AUROC) curve analysis. The features were correlated DXA and FRAX score.

Result: A set of three independent radiomic features (Dependence Non-Uniformity (DNU), Low Gray Level Emphasis (LGLE) and Kurtosis) showed significant ability to predict fragility fracture (AUROC DNU = 0.751, p < 0.05; AUROC LGLE = 0.729, p < 0.05; AUROC Kurtosis = 0.718, p < 0.05) with low to moderate correlation with FRAX and DXA.

Conclusion: Radiomic features can measure bone health in MRI of proximal femur and has the potential to predict fracture.