An editorial by Editor-in-Chief Maryellen Giger explains the journal’s transition to structured abstracts.

List of reviewers who served JMI in 2019.

The editorial introduces the Special Section on Evaluation Methodologies for Clinical AI.

Knowledge of the principles of image science is essential to the successful application of artificial intelligence in medical imaging.

We evaluated whether using synthetic mammograms for training data augmentation may reduce the effects of overfitting and increase the performance of a deep learning algorithm for breast mass detection. Synthetic mammograms were generated using in silico procedural analytic breast and breast mass modeling algorithms followed by simulated x-ray projections of the breast models into mammographic images. In silico breast phantoms containing masses were modeled across the four BI-RADS breast density categories, and the masses were modeled with different sizes, shapes, and margins. A Monte Carlo-based x-ray transport simulation code, MC-GPU, was used to project the three-dimensional phantoms into realistic synthetic mammograms. 2000 mammograms with 2522 masses were generated to augment a real data set during training. From the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) data set, we used 1111 mammograms (1198 masses) for training, 120 mammograms (120 masses) for validation, and 361 mammograms (378 masses) for testing. We used faster R-CNN for our deep learning network with pretraining from ImageNet using the Resnet-101 architecture. We compared the detection performance when the network was trained using different percentages of the real CBIS-DDSM training set (100%, 50%, and 25%), and when these subsets of the training set were augmented with 250, 500, 1000, and 2000 synthetic mammograms. Free-response receiver operating characteristic (FROC) analysis was performed to compare performance with and without the synthetic mammograms. We generally observed an improved test FROC curve when training with the synthetic images compared to training without them, and the amount of improvement depended on the number of real and synthetic images used in training. Our study shows that enlarging the training data with synthetic samples can increase the performance of deep learning systems.

The role of Ki-67 index in determining the prognosis and management of gastroenteropancreatic neuroendocrine tumors (GEP-NETs) has become more important yet presents a challenging assessment dilemma. Although the precise method of Ki-67 index evaluation has not been standardized, several methods have been proposed, and each has its pros and cons. Our study proposes an imaging semiautomated informatics framework [semiautomated counting (SAC)] using the popular biomedical imaging tool “ImageJ” to quantify Ki-67 index of the GEP-NETs using camera-captured images of tumor hotspots. It aims to assist pathologists in achieving an accurate and rapid interpretation of Ki-67 index and better reproducibility of the results with minimal human interaction and calibration. Twenty cases of resected GEP-NETs with Ki-67 staining that had been done for diagnostic purposes have been randomly selected from the pathology archive. All of these cases were reviewed in a multidisciplinary cancer center between 2012 and 2019. For each case, the Ki-67 immunostained slide was evaluated and five camera-captured images at 40  ×   magnification were taken. Prints of images were used by three pathologists to manually count the tumor cells. The digital versions of the images were used for the semiautomated cell counting using ImageJ. Statistical analysis of the Ki-67 index correlation between the proposed method and the MC revealed strong agreement on all the cases evaluates (n  =  20), with an intraclass correlation coefficient of 0.993, “95% CI: 0.984 to 0.997.” The results obtained from the SAC are promising and demonstrate the capability of this methodology for the development of reproducible and accurate semiautomated quantitative pathological assessments. ImageJ features are investigated carefully and accurately fine-tuned to obtain the optimal sequence of steps that will accurately calculate Ki-67 index. SAC is able to accurately grade all the cases evaluated perfectly mating histopathologists’ manual grading, providing reliable and efficient solution for Ki-67 index assessment.

Tumor volume is a topic of interest for the prognostic assessment, treatment response evaluation, and staging of malignant pleural mesothelioma. Many mesothelioma patients present with, or develop, pleural fluid, which may complicate the segmentation of this disease. Deep convolutional neural networks (CNNs) of the two-dimensional U-Net architecture were trained for segmentation of tumor in the left and right hemithoraces, with the networks initialized through layers pretrained on ImageNet. Networks were trained on a dataset of 5230 axial sections from 154 CT scans of 126 mesothelioma patients. A test set of 94 CT sections from 34 patients, who all presented with both tumor and pleural effusion, in addition to a more general test set of 130 CT sections from 43 patients, were used to evaluate segmentation performance of the deep CNNs. The Dice similarity coefficient (DSC), average Hausdorff distance, and bias in predicted tumor area were calculated through comparisons with radiologist-provided tumor segmentations on the test sets. The present method achieved a median DSC of 0.690 on the tumor and effusion test set and achieved significantly higher performance on both test sets when compared with a previous deep learning-based segmentation method for mesothelioma.

Purpose: Pathologists rely on relevant clinical information, visual inspection of stained tissue slide morphology, and sophisticated molecular diagnostics to accurately infer the biological origin of secondary metastatic cancer. While highly effective, this process is expensive in terms of time and clinical resources. We seek to develop and evaluate a computer vision system designed to reasonably infer metastatic origin of secondary liver cancer directly from digitized histopathological whole slide images of liver biopsy.

Approach: We illustrate a two-stage deep learning approach to accomplish this task. We first train a model to identify spatially localized regions of cancerous tumor within digitized hematoxylin and eosin (H&E)-stained tissue sections of secondary liver cancer based on a pathologist’s annotation of several whole slide images. Then, a second model is trained to generate predictions of the cancers’ metastatic origin belonging to one of three distinct clinically relevant classes as confirmed by immunohistochemistry.

Results: Our approach achieves a classification accuracy of 90.2% in determining metastatic origin of whole slide images from a held-out test set, which compares favorably to an established clinical benchmark by three board-certified pathologists whose accuracies ranged from 90.2% to 94.1% on the same prediction task.

Conclusions: We illustrate the potential impact of deep learning systems to leverage morphological and structural features of H&E-stained tissue sections to guide pathological and clinical determination of the metastatic origin of secondary liver cancers.

Purpose: Radiomic features extracted from medical images acquired in different countries may demonstrate a batch effect. Thus, we investigated the effect of harmonization on a database of radiomic features extracted from dynamic contrast-enhanced magnetic resonance (DCE-MR) breast imaging studies of 3150 benign lesions and cancers collected from international datasets, as well as the potential of harmonization to improve classification of malignancy.

Approach: Eligible features were harmonized by category using the ComBat method. Harmonization effect on features was evaluated using the Davies–Bouldin index for degree of clustering between populations for both benign lesions and cancers. Performance in distinguishing between cancers and benign lesions was evaluated for each dataset using 10-fold cross validation with the area under the receiver operating characteristic curve (AUC) determined on the pre- and postharmonization sets of radiomic features in each dataset and a combined one. Differences in AUCs were evaluated for statistical significance.

Results: The Davies–Bouldin index increased by 27% for benign lesions and by 43% for cancers, indicating that the postharmonization features were more similar. Classification performance using postharmonization features performed better than that using preharmonization features (p  <  0.001 for all three).

Conclusion: Harmonization of radiomic features may enable combining databases from different populations for more comprehensive computer-aided diagnosis models of breast cancer.

X-ray cone-beam holotomography of unstained tissue from the human central nervous system reveals details down to subcellular length scales. This visualization of variations in the electron density of the sample is based on phase-contrast techniques using intensities formed by self-interference of the beam between object and detector. Phase retrieval inverts diffraction and overcomes the phase problem by constraints such as several measurements at different Fresnel numbers for a single projection. Therefore, the object-to-detector distance (defocus) can be varied. However, for cone-beam geometry, changing defocus changes magnification, which can be problematic in view of image processing and resolution. Alternatively, the photon energy can be altered (multi-E). Far from absorption edges, multi-E data yield the wavelength-independent electron density. We present the multi-E holotomography at the Göttingen Instrument for Nano-Imaging with X-Rays (GINIX) setup of the P10 beamline at Deutsches Elektronen-Synchrotron. The instrument is based on a combined optics of elliptical mirrors and an x-ray waveguide positioned in the focal plane for further coherence, spatial filtering, and high numerical aperture. Previous results showed the suitability of this instrument for nanoscale tomography of unstained brain tissue. We demonstrate that upon energy variation, the focal spot is stable enough for imaging. To this end, a double-crystal monochromator and automated alignment routines are required. Three tomograms of human brain tissue were recorded and jointly analyzed using phase retrieval based on the contrast transfer function formalism generalized to multiple photon energies. Variations of the electron density of the sample are successfully reconstructed.

Purpose: Recently, progress has been achieved in implementing phase-contrast tomography of soft biological tissues at laboratory sources. This opens up opportunities for three-dimensional (3-D) histology based on x-ray computed tomography (μ- and nanoCT) in the direct vicinity of hospitals and biomedical research institutions. Combining advanced x-ray generation and detection techniques with phase reconstruction algorithms, 3-D histology can be obtained even of unstained tissue of the central nervous system, as shown, for example, for biopsies and autopsies of human cerebellum. Depending on the setup, i.e., source, detector, and geometric parameters, laboratory-based tomography can be implemented at very different sizes and length scales.

We investigate the extent to which 3-D histology of neuronal tissue can exploit the cone-beam geometry at high magnification M using a nanofocus transmission x-ray tube (nanotube) with a 300 nm minimal spot size (Excillum), combined with a single-photon counting camera. Tightly approaching the source spot with the biopsy punch, we achieve high M  ≈  10¹  −  10², high flux density, and exploit the superior efficiency of this detector technology.

Approach: Different nanotube configurations such as spot size and flux, M, as well as exposure time, Fresnel number, and coherence are varied and selected in view of resolution, field of view, and/or phase-contrast requirements.

Results: The data show that the information content for the cytoarchitecture is enhanced by the phase effect. Comparison of results to those obtained at a microfocus rotating-anode x-ray tomography setup with a high-resolution detector, i.e., in low-M geometry, reveals similar to slightly superior data quality for the nanotube setup. In addition to its compactness, reduced power consumption by a factor of 10³, and shorter scan duration, the particular advantage of the nanotube setup also lies in its suitability for pixel detector technology, enabling an increased range of opportunities for applications in laboratory phase-contrast x-ray tomography.

Conclusions: The phase retrieval scheme utilized mixes amplitude and phase contrast, with results being robust with respect to reconstruction parameters. Structural information content is comparable to slightly superior to previous results achieved with a microfocus rotating-anode setup but can be obtained in shorter scan time. Beyond advantages as compactness, lowered power consumption, and flexibility, the nanotube setup’s scalability in view of the progress in pixel detector technology is particularly beneficial. Further progress is thus likely to bring 3-D virtual histology to the performance in scan time and throughput required for clinical practice in neuropathology.

Several autofocus algorithms based on the analysis of image sharpness have been proposed for microscopy applications. Since autofocus functions (AFs) are computed from several images captured at different lens positions, these algorithms are considered computationally intensive. With the aim of presenting the capabilities of dedicated hardware to speed-up the autofocus process, we discuss the implementation of four AFs using, respectively, a multicore central processing unit (CPU) architecture and a graphic processing unit (GPU) card. Throughout different experiments performed on 300 image stacks previously identified with tuberculosis bacilli, the proposed implementations have allowed for the acceleration of the computation time for some AFs up to 23 times with respect to the serial version. These results show that the optimal use of multicore CPU and GPUs can be used effectively for autofocus in real-time microscopy applications.

Compressed sensing is an acquisition strategy that possesses great potential to accelerate magnetic resonance imaging (MRI) within the ambit of existing hardware, by enforcing sparsity on MR image slices. Compared to traditional reconstruction methods, dictionary learning-based reconstruction algorithms, which locally sparsify image patches, have been found to boost the reconstruction quality. However, due to the learning complexity, they have to be independently employed on successive MR undersampled slices one at a time. This causes them to forfeit prior knowledge of the anatomical structure of the region of interest. An MR reconstruction algorithm is proposed that employs the double sparsity model coupled with online sparse dictionary learning to learn directional features of the region under observation from existing prior knowledge. This is found to enhance the capability of sparsely representing directional features in an MR image and results in better reconstructions. The proposed framework is shown to have superior performance compared to state-of-art MRI reconstruction algorithms under noiseless and noisy conditions for various undersampling percentages and distinct scanning strategies.

Purpose: The pattern of dense tissue on a mammogram appears to provide additional information than overall density for risk assessment, but there has been little consistency in measures of texture identified. The purpose of this study is thus to validate a mammographic texture feature developed from a previous study in a new setting.

Approach: A case–control study (316 invasive cases and 1339 controls) of women in Virginia, USA was used to validate a mammographic texture feature (MMTEXT) derived in a independent previous study. Analysis of predictive ability was adjusted for age, demographic factors, questionnaire risk factors (combined through the Tyrer-Cuzick model), and optionally BI-RADS breast density. Odds ratios per interquartile range (IQ-OR) in controls were estimated. Subgroup analysis assessed heterogeneity by mode of cancer detection (94 not detected by mammography).

Results: MMTEXT was not a significant risk factor at 0.05 level after adjusting for classical risk factors (IQ-OR  =  1.16, 95%CI 0.92 to 1.46), nor after further adjustment for BI-RADS density (IQ-OR  =  0.92, 95%CI 0.76 to 1.10). There was weak evidence that MMTEXT was more predictive for cancers that were not detected by mammography (unadjusted for density: IQ-OR  =  1.46, 95%CI 0.99 to 2.15 versus 1.03, 95%CI 0.79 to 1.35, Phet 0.10; adjusted for density: IQ-OR  =  1.11, 95%CI 0.70 to 1.77 versus 0.76, 95%CI 0.55 to 1.05, Phet 0.21).

Conclusions: MMTEXT is unlikely to be a useful imaging marker for invasive breast cancer risk assessment in women attending mammography screening. Future studies may benefit from a larger sample size to confirm this as well as developing and validating other measures of risk. This negative finding demonstrates the importance of external validation.

Purpose: Placental size in early pregnancy has been associated with important clinical outcomes, including fetal growth. However, extraction of placental size from three-dimensional ultrasound (3DUS) requires time-consuming interactive segmentation methods and is prone to user variability. We propose a semiautomated segmentation technique that requires minimal user input to robustly measure placental volume from 3DUS images.

Approach: For semiautomated segmentation, a single, central 2D slice was manually annotated to initialize an automated multi-atlas label fusion (MALF) algorithm. The dataset consisted of 47 3DUS volumes obtained at 11 to 14 weeks in singleton pregnancies (28 anterior and 19 posterior). Twenty-six of these subjects were imaged twice within the same session. Dice overlap and surface distance were used to quantify the automated segmentation accuracy compared to expert manual segmentations. The mean placental volume measurements obtained by our method and VOCAL (virtual organ computer-aided analysis), a leading commercial semiautomated method, were compared to the manual reference set. The test–retest reliability was also assessed.

Results: The overlap between our automated segmentation and manual (mean Dice: 0.824  ±  0.061, median: 0.831) was within the range reported by other methods requiring extensive manual input. The average surface distance was 1.66  ±  0.96  mm. The correlation coefficient between test–retest volumes was r  =  0.88, and the intraclass correlation was ICC  (  1  )    =  0.86.

Conclusions: MALF is a promising method that can allow accurate and reliable segmentation of the placenta with minimal user interaction. Further refinement of this technique may allow for placental biometry to be incorporated into clinical pregnancy surveillance.

Purpose: Voxel-level hypothesis testing on images suffers from test multiplicity. Numerous correction methods exist, mainly applied and evaluated on neuroimaging and synthetic datasets. However, newly developed approaches like Imiomics, using different data and less common analysis types, also require multiplicity correction for more reliable inference. To handle the multiple comparisons in Imiomics, we aim to evaluate correction methods on whole-body MRI and correlation analyses, and to develop techniques specifically suited for the given analyses.

Approach: We evaluate the most common familywise error rate (FWER) limiting procedures on whole-body correlation analyses via standard (synthetic no-activation) nominal error rate estimation as well as smaller prior-knowledge based stringency analysis. Their performance is compared to our anatomy-based method extensions.

Results: Results show that nonparametric methods behave better for the given analyses. The proposed prior-knowledge based evaluation shows that the devised extensions including anatomical priors can achieve the same power while keeping the FWER closer to the desired rate.

Conclusions: Permutation-based approaches perform adequately and can be used within Imiomics. They can be improved by including information on image structure. We expect such method extensions to become even more relevant with new applications and larger datasets.

Obstetric ultrasound is a fundamental ingredient of modern prenatal care with many applications including accurate dating of a pregnancy, identifying pregnancy-related complications, and diagnosis of fetal abnormalities. However, despite its many benefits, two factors currently prevent wide-scale uptake of this technology for point-of-care clinical decision-making in low- and middle-income country (LMIC) settings. First, there is a steep learning curve for scan proficiency, and second, there has been a lack of easy-to-use, affordable, and portable ultrasound devices. We introduce a framework toward addressing these barriers, enabled by recent advances in machine learning applied to medical imaging. The framework is designed to be realizable as a point-of-care ultrasound (POCUS) solution with an affordable wireless ultrasound probe, a smartphone or tablet, and automated machine-learning-based image processing. Specifically, we propose a machine-learning-based algorithm pipeline designed to automatically estimate the gestational age of a fetus from a short fetal ultrasound scan. We present proof-of-concept evaluation of accuracy of the key image analysis algorithms for automatic head transcerebellar plane detection, automatic transcerebellar diameter measurement, and estimation of gestational age on conventional ultrasound data simulating the POCUS task and discuss next steps toward translation via a first application on clinical ultrasound video from a low-cost ultrasound probe.

Planning of radiotherapy involves accurate segmentation of a large number of organs at risk (OAR), i.e., organs for which irradiation doses should be minimized to avoid important side effects of the therapy. We propose a deep learning method for segmentation of OAR inside the head, from magnetic resonance images (MRIs). Our system performs segmentation of eight structures: eye, lens, optic nerve, optic chiasm, pituitary gland, hippocampus, brainstem, and brain. We propose an efficient algorithm to train neural networks for an end-to-end segmentation of multiple and nonexclusive classes, addressing problems related to computational costs and missing ground truth segmentations for a subset of classes. We enforce anatomical consistency of the result in a postprocessing step. In particular, we introduce a graph-based algorithm for segmentation of the optic nerves, enforcing the connectivity between the eyes and the optic chiasm. We report cross-validated quantitative results on a database of 44 contrast-enhanced T1-weighted MRIs with provided segmentations of the considered OAR, which were originally used for radiotherapy planning. In addition, the segmentations produced by our model on an independent test set of 50 MRIs were evaluated by an experienced radiotherapist in order to qualitatively assess their accuracy. The mean distances between produced segmentations and the ground truth ranged from 0.1 to 0.7 mm across different organs. A vast majority (96%) of the produced segmentations were found acceptable for radiotherapy planning.

We are developing automated analysis of corneal-endothelial-cell-layer, specular microscopic images so as to determine quantitative biomarkers indicative of corneal health following corneal transplantation. Especially on these images of varying quality, commercial automated image analysis systems can give inaccurate results, and manual methods are very labor intensive. We have developed a method to automatically segment endothelial cells with a process that included image flattening, U-Net deep learning, and postprocessing to create individual cell segmentations. We used 130 corneal endothelial cell images following one type of corneal transplantation (Descemet stripping automated endothelial keratoplasty) with expert-reader annotated cell borders. We obtained very good pixelwise segmentation performance (e.g., Dice coefficient  =  0.87  ±  0.17, Jaccard index  =  0.80  ±  0.18, across 10 folds). The automated method segmented cells left unmarked by analysts and sometimes segmented cells differently than analysts (e.g., one cell was split or two cells were merged). A clinically informative visual analysis of the held-out test set showed that 92% of cells within manually labeled regions were acceptably segmented and that, as compared to manual segmentation, automation added 21% more correctly segmented cells. We speculate that automation could reduce 15 to 30 min of manual segmentation to 3 to 5 min of manual review and editing.

Purpose: While radiomics feature values can differ when extracted using different radiomics software, the effects of these variations when applied to a particular clinical task are currently unknown. The goal of our study was to use various radiomics software packages to classify patients with radiation pneumonitis (RP) and to quantify the variation in classification ability among packages.

Approach: A database of serial thoracic computed tomography scans was obtained from 105 patients with esophageal cancer. Patients were treated with radiation therapy (RT), resulting in 20 patients developing RP grade ≥2. Regions of interest (ROIs) were randomly placed in the lung volume of the pre-RT scan within high-dose regions (≥30  Gy), and corresponding ROIs were anatomically matched in the post-RT scan. Three radiomics packages were compared: A1 (in-house), IBEX v1.0 beta, and PyRadiomics v.2.0.0. Radiomics features robust to deformable registration and common among radiomics packages were calculated: four first-order and four gray-level co-occurrence matrix features. Differences in feature values between time points were calculated for each feature, and logistic regression was used in conjunction with analysis of variance to classify patients with and without RP (p  <  0.006). Classification ability for each package was assessed using receiver operating characteristic (ROC) analysis and compared using the area under the ROC curve (AUC).

Results: Of the eight radiomics features, five were significantly correlated with RP status for all three packages, whereas one feature was not significantly correlated with RP for all three packages. The remaining two features differed in whether or not they were significantly associated with RP status among the packages. Seven of the eight features agreed among the packages in whether the AUC value was significantly >0.5.

Conclusions: Radiomics features extracted using different software packages can result in differences in classification ability.

Performing large-scale three-dimensional radiation dose reconstruction for patients requires a large amount of manual work. We present an image processing-based pipeline to automatically reconstruct radiation dose. The pipeline was designed for childhood cancer survivors that received abdominal radiotherapy with anterior-to-posterior and posterior-to-anterior field set-up. First, anatomical landmarks are automatically identified on two-dimensional radiographs. Second, these landmarks are used to derive parameters to emulate the geometry of the plan on a surrogate computed tomography. Finally, the plan is emulated and used as input for dose calculation. For qualitative evaluation, 100 cases of automatic and manual plan emulations were assessed by two experienced radiation dosimetrists in a blinded comparison. The two radiation dosimetrists approved 100%/100% and 92%/91% of the automatic/manual plan emulations, respectively. Similar approval rates of 100% and 94% hold when the automatic pipeline is applied on another 50 cases. Further, quantitative comparisons resulted in on average <5  mm difference in plan isocenter/borders, and <0.9  Gy in organ mean dose (prescribed dose: 14.4 Gy) calculated from the automatic and manual plan emulations. No statistically significant difference in terms of dose reconstruction accuracy was found for most organs at risk. Ultimately, our automatic pipeline results are of sufficient quality to enable effortless scaling of dose reconstruction data generation.

Purpose: We describe a shape-aware multisurface simplex deformable model for the segmentation of healthy as well as pathological lumbar spine in medical image data.

Approach: This model provides an accurate and robust segmentation scheme for the identification of intervertebral disc pathologies to enable the minimally supervised planning and patient-specific simulation of spine surgery, in a manner that combines multisurface and shape statistics-based variants of the deformable simplex model. Statistical shape variation within the dataset has been captured by application of principal component analysis and incorporated during the segmentation process to refine results. In the case where shape statistics hinder detection of the pathological region, user assistance is allowed to disable the prior shape influence during deformation.

Results: Results demonstrate validation against user-assisted expert segmentation, showing excellent boundary agreement and prevention of spatial overlap between neighboring surfaces. This section also plots the characteristics of the statistical shape model, such as compactness, generalizability and specificity, as a function of the number of modes used to represent the family of shapes. Final results demonstrate a proof-of-concept deformation application based on the open-source surgery simulation Simulation Open Framework Architecture toolkit.

Conclusions: To summarize, we present a deformable multisurface model that embeds a shape statistics force, with applications to surgery planning and simulation.

We assessed interventional radiologists’ task-based image quality preferences for two- and three-dimensional images obtained with a complementary metal–oxide semiconductor (CMOS) flat-panel detector versus a hydrogenated amorphous silicon (a-Si:H) flat-panel detector. CMOS and a-Si:H detectors were implemented on identical mobile C-arms to acquire radiographic, fluoroscopic, and cone-beam computed tomography (CBCT) images of cadavers undergoing simulated interventional procedures using low- and high-dose settings. Images from both systems were displayed side by side on calibrated, diagnostic-quality displays, and three interventional radiologists evaluated task performance relevant to each image and ranked their preferences based on visibility of pertinent anatomy and interventional devices. Overall, CMOS images were preferred in fluoroscopy (p  =  0.002) and CBCT (p  =  0.004), at low-dose settings (p  =  0.001), and for tasks associated with high levels of spatial resolution [e.g., fine anatomical details (p  =  0.006) and assessment of interventional devices (p  =  0.015)]. No significant difference was found for fluoroscopic imaging tasks emphasizing temporal resolution (p  =  0.072), for radiography tasks (p  =  0.825), when using high-dose settings (p  =  0.360), or tasks involving general anatomy (p  =  0.174). The image quality preferences are consistent with reported technical advantages of CMOS regarding finer pixel size and reduced electronic noise.

Purpose: Biomarkers related to hemodynamics can be quantified using angiographic parametric imaging (API), which is a quantitative imaging method that uses digital subtraction angiography (DSA). We aimed to assess the accuracy of API in locating infarct core within large vessel occlusion (LVO) acute ischemic stroke (AIS) patients.

Approach: Data were retrospectively collected for 25 LVO AIS patients who achieved successful recanalization. DSA data from lateral and anteroposterior (AP) views were loaded into API software to generate hemodynamic parameter maps. Relative differences in hemispherical regions for each API parameter were calculated. Ground truth infarct core locations were obtained using 24-h follow-up fluid-attenuation inversion recovery (FLAIR) MRI imaging. FLAIR MRI infarct locations were registered with DSA images to determine infarct regions in API parameter maps. Relative differences across hemispheres for each API parameter were plotted against each other. A support vector machine was used to determine the optimal hyperplane for classifying regions as infarct or healthy tissue.

Results: For the lateral and AP views, respectively, the most accurate classification of infarct regions came from plotting mean transit time (MTT) versus peak height (PH) [accuracy  =  0.8125  ±  0.0012 (95%)], the area under the receiver operator characteristic curve   (  AUROC  )    =  0.8946  ±  0.0000 (95%), and plotting MTT versus the area under the curve (AUC) [accuracy  =  0.7957  ±  0.0011 (95%), AUROC  =  0.8759  ±  0.0000 (95%)].

Conclusions: API provides accurate assessment of locating ischemic core in AIS LVO patients and has the potential for clinical benefit by determining infarct core location and growth in real time for intraoperative decision making.

Alzheimer’s disease (AD) is characterized by the progressive accumulation of neurofibrillary tangles associated with amyloid plaques. We used 80 resting-state functional magnetic resonance imaging and 80 T₁ images acquired using MP-RAGE (magnetization-prepared rapid acquisition gradient echo) from Alzheimer’s Disease Neuroimaging Initiative data to detect atrophy changes and functional connectivity patterns of the default mode networks (DMNs). The study subjects were classified into four groups (each with n  =  20) based on their Mini-Mental State Examination (MMSE) score as follows: cognitively normal (CN), early mild cognitive impairment, late mild cognitive impairment, and AD. The resting-state functional connectivity of the DMN was examined between the groups using the CONN functional connectivity toolbox. Loss of gray matter in AD was observed. Atrophy measured by the volume of selected subcortical regions, using the Functional Magnetic Resonance Imaging of the Brain (FMRIB) Software Library’s Integrated Registration and Segmentation Tool (FIRST), revealed significant volume loss in AD when compared to CN (p  <  0.05). DMNs were selected to assess functional connectivity. The negative connectivity of DMN increased in AD group compared to controls. Graph theory parameters, such as global and local efficiency, betweenness centrality, average path length, and cluster coefficient, were computed. Relatively higher correlation between MMSE and functional metrics (r  =  0.364, p  =  0.001) was observed as compared to atrophy measures (r  =  0.303, p  =  0.006). In addition, the receiver operating characteristic analysis showed large area under the curve (A_Z) for functional parameters (A_Z  >  0.9), compared to morphometric changes (A_Z  <  0.8). In summary, it is observed that the functional connectivity measures may serve a better predictor in comparison to structural atrophy changes. We postulate that functional connectivity measures have the potential to evolve as a marker for the early detection of AD.

DICOM header information is frequently used to classify medical image types; however, if a header is missing fields or contains incorrect data, the utility is limited. To expedite image classification, we trained convolutional neural networks (CNNs) in two classification tasks for thoracic radiographic views obtained from dual-energy studies: (a) distinguishing between frontal, lateral, soft tissue, and bone images and (b) distinguishing between posteroanterior (PA) or anteroposterior (AP) chest radiographs. CNNs with AlexNet architecture were trained from scratch. 1910 manually classified radiographs were used for training the network to accomplish task (a), then tested with an independent test set (3757 images). Frontal radiographs from the two datasets were combined to train a network to accomplish task (b); tested using an independent test set of 1000 radiographs. ROC analysis was performed for each trained CNN with area under the curve (AUC) as a performance metric. Classification between frontal images (AP/PA) and other image types yielded an AUC of 0.997 [95% confidence interval (CI): 0.996, 0.998]. Classification between PA and AP radiographs resulted in an AUC of 0.973 (95% CI: 0.961, 0.981). CNNs were able to rapidly classify thoracic radiographs with high accuracy, thus potentially contributing to effective and efficient workflow.

We present a roadmap for integrating artificial intelligence (AI)-based image analysis algorithms into existing radiology workflows such that (1) radiologists can significantly benefit from enhanced automation in various imaging tasks due to AI, and (2) radiologists’ feedback is utilized to further improve the AI application. This is achieved by establishing three maturity levels where (1) research enables the visualization of AI-based results/annotations by radiologists without generating new patient records; (2) production allows the AI-based system to generate results stored in an institution’s picture-archiving and communication system; and (3) feedback equips radiologists with tools for editing the AI inference results for periodic retraining of the deployed AI systems, thereby allowing continuous organic improvement of AI-based radiology-workflow solutions. A case study (i.e., detection of brain metastases with T1-weighted contrast-enhanced three-dimensional MRI) illustrates the deployment details of a particular AI-based application according to the aforementioned maturity levels. It is shown that the given AI application significantly improves with feedback coming from radiologists; the number of incorrectly detected brain metastases (false positives) decreases from 14.2 to 9.12 per patient with the number of subsequently annotated datasets increasing from 93 to 217 as a result of radiologist adjudication.