Purpose: Flat-panel radiography detectors employ thin-film transistor (TFT) panels to acquire high-quality x-ray images. Pixel defects occur due to circuit shorts or opens in the TFT panel. The defects may degrade the image quality, as well as lower the production yield, and eventually raise the production cost. Hence, it is important to develop an appropriate defect correction algorithm for acquired images. Traditional correction algorithms are based on a complicated adaptive filtering technique, which exploits neighbor pixels, to faithfully preserve the edge components. Because of the complexity of the traditional sophisticated approaches, optimizing their correction performances is difficult.

Approach: We considered various pixel-defect correction algorithms based on different deep learning models, such as the artificial neural network (ANN), convolutional neural network (CNN), concatenate CNN, and generative adversarial networks (GAN). We considered two cases of maximal defect sizes, 3  ×  3 and 5  ×  5  pixels, and conducted extensive learning experiments to find the best structures of the learning models using the mean square error (MSE) as the loss function.

Results: To conduct experiments, practical chest x-ray images were acquired from a general radiography detector. The MSE values of the correction results from ANN, CNN, concatenate CNN, and GAN were 69.40, 75.13, 68.21, and 73.77, respectively, and were much smaller than that of the conventional template match correction method.

Conclusions: A concatenate CNN showed the best defect-correction performance. However, ANN could achieve a similar correction performance with much smaller encoding complexity. Therefore, the single-layer ANN can efficiently conduct defect corrections in terms of both correction and complexity.

Purpose: The focal spot size and shape of an x-ray system are critical factors to the spatial resolution. Conventional approaches to characterizing the focal spot use specialized tools that usually require careful calibration. We propose an alternative to characterize the x-ray source’s focal spot, simply using a rotating edge and flat-panel detector.

Methods: An edge is moved to the beam axis, and an edge spread function (ESF) is obtained at a specific angle. Taking the derivative of the ESF provides the line spread function, which is the Radon transform of the focal spot in the direction parallel to the edge. By rotating the edge about the beam axis for 360 deg, we obtain a complete Radon transform, which is used for reconstructing the focal spot. We conducted a study on a clinical C-arm system with three focal spot sizes (0.3, 0.6, and 1.0 mm nominal size), then compared the focal spot imaged using the proposed method against the conventional pinhole approach. The full width at half maximum (FWHM) of the focal spots along the width and height of the focal spot were used for quantitative comparisons.

Results: Using the pinhole method as ground truth, the proposed method accurately characterized the focal spot shapes and sizes. Quantitatively, the FWHM widths were 0.37, 0.65, and 1.14 mm for the pinhole method and 0.33, 0.60, and 1.15 mm for the proposed method for the 0.3, 0.6, and 1.0 mm nominal focal spots, respectively. Similar levels of agreement were found for the FWHM heights.

Conclusions: The method uses a rotating edge to characterize the focal spot and could be automated in the future using a system’s built-in collimator. The method could be included as part of quality assurance tests of image quality and tube health.

Purpose: In sequential imaging studies, there exists rich information from past studies that can be used in prior-image-based reconstruction (PIBR) as a form of improved regularization to yield higher-quality images in subsequent studies. PIBR methods, such as reconstruction of difference (RoD), have demonstrated great improvements in the image quality of subsequent anatomy reconstruction even when CT data are acquired at very low-exposure settings.

Approach: However, to effectively use information from past studies, two major elements are required: (1) registration, usually deformable, must be applied between the current and prior scans. Such registration is greatly complicated by potential ambiguity between patient motion and anatomical change—which is often the target of the followup study. (2) One must select regularization parameters for reliable and robust reconstruction of features.

Results: We address these two major issues and apply a modified RoD framework to the clinical problem of lung nodule surveillance. Specifically, we develop a modified deformable registration approach that enforces a locally smooth/rigid registration around the change region and extend previous analytic expressions relating reconstructed contrast to the regularization parameter and other system dependencies for reliable representation of image features. We demonstrate the efficacy of this approach using a combination of realistic digital phantoms and clinical projection data. Performance is characterized as a function of the size of the locally smooth registration region of interest as well as x-ray exposure.

Conclusions: This modified framework is effectively able to separate patient motion and anatomical change to directly highlight anatomical change in lung nodule surveillance.

Purpose: The reconstruction of positron emission tomography images is a computationally intensive task which benefits from the use of increasingly complex physical models. Aiming to reduce the computational burden by means of a reduced system matrix, we present a list mode reconstruction approach based on maximum likelihood-expectation maximization and a sliced mesh support.

Approach: The reconstruction strategy uses a fully 3D projection along series of 2D meshes arranged in the axial plane of the scanner. These series of meshes describe the continuous volumetric activity using a piece-wise linear function interpolated from the mesh elements. The mesh support is automatically adapted to the underlying structure of the activity by means of a remeshing process. This process finds a high-quality compact mesh representation constrained to a controlled interpolation error.

Results: The method is tested using a Monte Carlo simulation of a Hoffman brain phantom and a National Electrical Manufacturers Association image quality phantom acquisition, using different sets of statistics. The reconstructions are compared against a voxelized reconstruction under different conditions, achieving similar or superior results. The number of parameters needed to reconstruct the image in voxel and mesh support is also compared, and the mesh reconstruction permits to reduce the number of nodes used to represent a complex image.

Conclusions: The proposed reconstruction strategy reduces the number of parameters needed to describe the activity distribution by more than one order of magnitude for similar voxel size and with similar accuracy than state-of-the-art methods.

Purpose: Automatic and consistent meningioma segmentation in T1-weighted magnetic resonance (MR) imaging volumes and corresponding volumetric assessment is of use for diagnosis, treatment planning, and tumor growth evaluation. We optimized the segmentation and processing speed performances using a large number of both surgically treated meningiomas and untreated meningiomas followed at the outpatient clinic.

Approach: We studied two different three-dimensional (3D) neural network architectures: (i) a simple encoder-decoder similar to a 3D U-Net, and (ii) a lightweight multi-scale architecture [Pulmonary Lobe Segmentation Network (PLS-Net)]. In addition, we studied the impact of different training schemes. For the validation studies, we used 698 T1-weighted MR volumes from St. Olav University Hospital, Trondheim, Norway. The models were evaluated in terms of detection accuracy, segmentation accuracy, and training/inference speed.

Results: While both architectures reached a similar Dice score of 70% on average, the PLS-Net was more accurate with an F1-score of up to 88%. The highest accuracy was achieved for the largest meningiomas. Speed-wise, the PLS-Net architecture tended to converge in about 50 h while 130 h were necessary for U-Net. Inference with PLS-Net takes less than a second on GPU and about 15 s on CPU.

Conclusions: Overall, with the use of mixed precision training, it was possible to train competitive segmentation models in a relatively short amount of time using the lightweight PLS-Net architecture. In the future, the focus should be brought toward the segmentation of small meningiomas (<2  ml) to improve clinical relevance for automatic and early diagnosis and speed of growth estimates.

Purpose: In clinical practice, positron emission tomography (PET) images are mostly analyzed visually, but the sensitivity and specificity of this approach greatly depend on the observer’s experience. Quantitative analysis of PET images would alleviate this problem by helping define an objective limit between normal and pathological findings. We present an anomaly detection framework for the individual analysis of PET images.

Approach: We created subject-specific abnormality maps that summarize the pathology’s topographical distribution in the brain by comparing the subject’s PET image to a model of healthy PET appearance that is specific to the subject under investigation. This model was generated from demographically and morphologically matched PET scans from a control dataset.

Results: We generated abnormality maps for healthy controls, patients at different stages of Alzheimer’s disease and with different frontotemporal dementia syndromes. We showed that no anomalies were detected for the healthy controls and that the anomalies detected from the patients with dementia coincided with the regions where abnormal uptake was expected. We also validated the proposed framework using the abnormality maps as inputs of a classifier and obtained higher classification accuracies than when using the PET images themselves as inputs.

Conclusions: The proposed method was able to automatically locate and characterize the areas characteristic of dementia from PET images. The abnormality maps are expected to (i) help clinicians in their diagnosis by highlighting, in a data-driven fashion, the pathological areas, and (ii) improve the interpretability of subsequent analyses, such as computer-aided diagnosis or spatiotemporal modeling.

Purpose: Sharing medical images between institutions, or even inside the same institution, is restricted by various laws and regulations; research projects requiring large datasets may suffer as a result. These limitations might be addressed by an abundant supply of synthetic data that (1) are representative (i.e., the synthetic data could produce comparable research results as the original data) and (2) do not closely resemble the original images (i.e., patient privacy is protected). We introduce a framework that generates data with these requirements leveraging generative adversarial network (GAN) ensembles in a controlled fashion.

Approach: To this end, an adaptive ensemble scaling strategy with the objective of representativeness is defined. A sampled Fréchet distance-based constraint was then created to eliminate poorly converged candidates. Finally, a mutual information-based validation metric was embedded into the framework to confirm there are visual differences between the original and the generated synthetic images.

Results: The applicability of the solution is demonstrated with a case study for generating three-dimensional brain metastasis (BM) from T1-weighted contrast-enhanced MRI studies. A previously published BM detection system was reported to produce 9.12 false-positives at 90% detection sensitivity based on the original data. By using the synthetic data generated with the proposed framework, the system produced 9.53 false-positives at the same sensitivity level.

Conclusions: Achieving comparable algorithm performance relying solely on synthetic data unveils a significant potential to eliminate/reduce patient privacy concerns when sharing data in medical imaging.

Purpose: A computer-aided diagnosis (CADx) system for breast masses is proposed, which incorporates both handcrafted and convolutional radiomic features embedded into a single deep learning model.

Approach: The model combines handcrafted and convolutional radiomic signatures into a multi-view architecture, which retrieves three-dimensional (3D) image information by simultaneously processing multiple two-dimensional mass patches extracted along different planes through the 3D mass volume. Each patch is processed by a stream composed of two concatenated parallel branches: a multi-layer perceptron fed with automatically extracted handcrafted radiomic features, and a convolutional neural network, for which discriminant features are learned from the input patches. All streams are then concatenated together into a final architecture, where all network weights are shared and the learning occurs simultaneously for each stream and branch. The CADx system was developed and tested for diagnosis of breast masses (N  =  284) using image datasets acquired with independent dedicated breast computed tomography systems from two different institutions. The diagnostic classification performance of the CADx system was compared against other machine and deep learning architectures adopting handcrafted and convolutional approaches, and three board-certified breast radiologists.

Results: On a test set of 82 masses (45 benign, 37 malignant), the proposed CADx system performed better than all other model architectures evaluated, with an increase in the area under the receiver operating characteristics curve (AUC) of 0.05  ±  0.02, and achieving a final AUC of 0.947, outperforming the three radiologists (AUC  =  0.814  −  0.902).

Conclusions: In conclusion, the system demonstrated its potential usefulness in breast cancer diagnosis by improving mass malignancy assessment.

Purpose: We present a markerless vision-based method for on-the-fly three-dimensional (3D) pose estimation of a fiberscope instrument to target pathologic areas in the endoscopic view during exploration.

Approach: A 2.5-mm-diameter fiberscope is inserted through the endoscope’s operating channel and connected to an additional camera to perform complementary observation of a targeted area such as a multimodal magnifier. The 3D pose of the fiberscope is estimated frame-by-frame by maximizing the similarity between its silhouette (automatically detected in the endoscopic view using a deep learning neural network) and a cylindrical shape bound to a kinematic model reduced to three degrees-of-freedom. An alignment of the cylinder axis, based on Plücker coordinates from the straight edges detected in the image, makes convergence faster and more reliable.

Results: The performance on simulations has been validated with a virtual trajectory mimicking endoscopic exploration and on real images of a chessboard pattern acquired with different endoscopic configurations. The experiments demonstrated a good accuracy and robustness of the proposed algorithm with errors of 0.33  ±  0.68  mm in distance position and 0.32  ±  0.11 deg in axis orientation for the 3D pose estimation, which reveals its superiority over previous approaches. This allows multimodal image registration with sufficient accuracy of <3  pixels.

Conclusion: Our pose estimation pipeline was executed on simulations and patterns; the results demonstrate the robustness of our method and the potential of fiber-optical instrument image-based tracking for pose estimation and multimodal registration. It can be fully implemented in software and therefore easily integrated into a routine clinical environment.

Purpose: Published reports suggest that nonoptimal visual search behavior is associated with false negatives in chest x-ray interpretation. Eye movement modeling example (EMME)-based training interventions, that is, interventions showing models of visual search to trainees, have been shown to improve visual search as well as task accuracy.

Approach: We examined the detection of focal lung pathology on chest x-rays before and after two different EMME training interventions that have been shown to be efficient: (i) an EMME showing moving fixations on a blurred background (spotlight group) and (ii) an EMME showing moving fixations on a nonblurred background (circle group). These two experimental groups were compared to a control group that was only provided with the correct location of pathologies on the chest x-rays.

Results: Performance outcomes showed improved detection sensitivity and specificity in all groups (also the control group). It appears that repetitive exposure to pathologies on chest x-rays with feedback resulted in enhanced pattern recognition. In addition, visual search strategies became more efficient during post-tests.

Conclusion: Repetitive exposure to a focal lung pathology detection task with feedback improves overall performance. However, the specific EMME training interventions did not add any further advantages. Similar training interventions can be provided online to assess feasibility and reproducibility of such (or similar) training formats. Such training can, for example, reduce the number of false negative errors, especially for novices.

Purpose: We report that a large field-of-view (FOV) retinal image can be acquired by a smart fundus camera.

Approach: This handheld system consists of a Raspberry Pi board, a touch screen display, a customized optical lens group, a ring light-emitting diode, and a Li-battery.

Results: Wide FOV of ∼57  deg is observed with proper lens configuration and can be expanded using image stitching algorithms.

Conclusions: This customized handheld fundus camera provides better image quality than cellphone-based fundus imaging solutions and offers more operational features than traditional portable fundus cameras. It may benefit field-portable ophthalmic diagnostic applications.

Purpose: Micro-computed tomography (micro-CT) scan provides high-resolution three-dimensional images of mineralized tissues in small animal models. Contrast enhancement is essential to visualize non-mineralized tissues with micro-CT scan. We attempted to compare the two most common contrast agents to stain and image mouse cardiac structures.

Approach: Ex-vivo micro-CT scan images of the mouse hearts were obtained following staining by potassium iodide or phosphotungstic acid (PTA). PTA-stained samples were imaged after various durations following staining (14 days, 25 days, 187 days, and 780 days), whereas iodine-stained samples were imaged after 72 hours. We compared median staining intensity between PTA and iodine at 0.1-mm intervals from the edge using the Mann Whitney test with correction for multiple comparisons.

Results: Sixty post-natal mice hearts were stained with either PTA or iodine and imaged using micro-CT scan. Iodine proved to be faster and more uniform in complete enhancement of cardiac tissue in as short as 72 h, whereas PTA required a significantly longer time period to penetrate mouse cardiac structure (>150  days). Median staining intensity with iodine was strongly higher than that with PTA from 0.1- to 1.5-mm distance from the epicardial edge (2-tailed P value <0.01 or lower throughout).

Conclusions: Iodine-stained soft tissue imaging by micro-CT scan provides a non-destructive, efficient, and accurate visualization tool for anatomical analysis of animal heart models of human cardiovascular conditions. Iodine is more efficient compared to PTA to achieve complete murine myocardial staining in a significantly shorter time period.

Purpose: The mean linear intercept (MLI) score is a common metric for quantification of injury in lung histopathology images. The automated estimation of the MLI score is a challenging task because it requires accurate segmentation of different biological components of the lung tissue. Therefore, the most widely used approaches for MLI quantification are based on manual/semi-automated assessment of lung histopathology images, which can be expensive and time-consuming. We describe a fully automated pipeline for MLI estimation, which is capable of producing results comparable to human raters.

Approach: We use a convolutional neural network based on U-Net architecture to segment the diagnostically relevant tissue segments in the whole slide images (WSI) of the mouse lung tissue. The proposed method extracts multiple field-of-view (FOV) images from the tissue segments and screen the FOV images, rejecting images based on presence of certain biological structures (i.e., blood vessels and bronchi). We used color slicing and region growing for segmentation of different biological structures in each FOV image.

Results: The proposed method was tested on ten WSIs from mice and compared against the scores provided by three human raters. In segmenting the relevant tissue segments, our method obtained a mean accuracy, Dice coefficient, and Hausdorff distance of 98.34%, 98.22%, and 109.68  μm, respectively. Our proposed method yields a mean precision, recall, and F1-score of 93.37%, 83.47%, and 87.87%, respectively, in screening of FOV images. There was substantial agreement found between the proposed method and the manual scores (Fleiss Kappa score of 0.76). The mean difference between the calculated MLI score between the automated method and average rater’s score was 2.33  ±  4.13 (4.25  %    ±  5.67  %  ).

Conclusion: The proposed pipeline for automated calculation of the MLI score demonstrates high consistency and accuracy with human raters and can be a potential replacement for manual/semi-automated approaches in the field.