Improved Cancer Detection Using Artificial Intelligence: a Retrospective Evaluation of Missed Cancers on Mammography.

To determine whether cmAssist™, an artificial intelligence-based computer-aided detection (AI-CAD) algorithm, can be used to improve radiologists' sensitivity in breast cancer screening and detection. A blinded retrospective study was performed with a panel of seven radiologists using a cancer-enriched data set from 122 patients that included 90 false-negative mammograms obtained up to 5.8 years prior to diagnosis and 32 BIRADS 1 and 2 patients with a 2-year follow-up of negative diagnosis. The mammograms were performed between February 7, 2008 (earliest) and January 8, 2016 (latest), and were all originally interpreted as negative in conjunction with R2 ImageChecker CAD, version 10.0. In this study, the readers analyzed the 122 studies before and after review of cmAssist™, an AI-CAD software for mammography. The statistical significance of our findings was evaluated using Student's t test and bootstrap statistical analysis. There was a substantial and significant improvement in radiologist accuracy with use of cmAssist, as demonstrated in the 7.2% increase in the area-under-the-curve (AUC) of the receiver operating characteristic (ROC) curve with two-sided p value < 0.01 for the reader group. All radiologists showed a significant improvement in their cancer detection rate (CDR) with the use of cmAssist (two-sided p value = 0.030, confidence interval = 95%). The readers detected between 25 and 71% (mean 51%) of the early cancers without assistance. With cmAssist, the overall reader CDR was 41 to 76% (mean 62%). The percentage increase in CDR for the reader panel was significant, ranging from 6 to 64% (mean 27%) with the use of cmAssist. There was less than 1% increase in the readers' false-positive recalls with use of cmAssist. With the use of cmAssist TM, there was a substantial and statistically significant improvement in radiologists' accuracy and sensitivity for detection of cancers that were originally missed. The percentage increase in CDR for the radiologists in the reader panel ranged from 6 to 64% (mean 27%) with the use of cmAssist, with negligible increase in false-positive recalls.

Brain swelling and death in children with cerebral malaria.

BACKGROUND: Case fatality rates among African children with cerebral malaria remain in the range of 15 to 25%. The key pathogenetic processes and causes of death are unknown, but a combination of clinical observations and pathological findings suggests that increased brain volume leading to raised intracranial pressure may play a role. Magnetic resonance imaging (MRI) became available in Malawi in 2009, and we used it to investigate the role of brain swelling in the pathogenesis of fatal cerebral malaria in African children. METHODS: We enrolled children who met a stringent definition of cerebral malaria (one that included the presence of retinopathy), characterized them in detail clinically, and obtained MRI scans on admission and daily thereafter while coma persisted. RESULTS: Of 348 children admitted with cerebral malaria (as defined by the World Health Organization), 168 met the inclusion criteria, underwent all investigations, and were included in the analysis. A total of 25 children (15%) died, 21 of whom (84%) had evidence of severe brain swelling on MRI at admission. In contrast, evidence of severe brain swelling was seen on MRI in 39 of 143 survivors (27%). Serial MRI scans showed evidence of decreasing brain volume in the survivors who had had brain swelling initially. CONCLUSIONS: Increased brain volume was seen in children who died from cerebral malaria but was uncommon in those who did not die from the disease, a finding that suggests that raised intracranial pressure may contribute to a fatal outcome. The natural history indicates that increased intracranial pressure is transient in survivors. (Funded by the National Institutes of Health and Wellcome Trust U.K.).

Vaccinia virus-mediated melanin production allows MR and optoacoustic deep tissue imaging and laser-induced thermotherapy of cancer.

We reported earlier the delivery of antiangiogenic single chain antibodies by using oncolytic vaccinia virus strains to enhance their therapeutic efficacy. Here, we provide evidence that gene-evoked production of melanin can be used as a therapeutic and diagnostic mediator, as exemplified by insertion of only one or two genes into the genome of an oncolytic vaccinia virus strain. We found that produced melanin is an excellent reporter for optical imaging without addition of substrate. Melanin production also facilitated deep tissue optoacoustic imaging as well as MRI. In addition, melanin was shown to be a suitable target for laser-induced thermotherapy and enhanced oncolytic viral therapy. In conclusion, melanin as a mediator for thermotherapy and reporter for different imaging modalities may soon become a versatile alternative to replace fluorescent proteins also in other biological systems. After ongoing extensive preclinical studies, melanin overproducing oncolytic virus strains might be used in clinical trials in patients with cancer.

ACR guidance document on MR safe practices: 2013.

Because there are many potential risks in the MR environment and reports of adverse incidents involving patients, equipment and personnel, the need for a guidance document on MR safe practices emerged. Initially published in 2002, the ACR MR Safe Practices Guidelines established de facto industry standards for safe and responsible practices in clinical and research MR environments. As the MR industry changes the document is reviewed, modified and updated. The most recent version will reflect these changes.

Assessment of tracer delay effect in whole-brain computed tomography perfusion: results for selected regions of interest in middle cerebral artery acute ischemic strokes.

OBJECTIVE: The objective of this study was to compare the variability of computed tomography perfusion (CTP) results in identical data sets of middle cerebral artery (MCA) acute ischemic stroke (AIS) generated by standard singular value decomposition (sSVD) deconvolution and tracer delay-insensitive singular value decomposition (SVD+) algorithm analyses. METHODS: Whole-brain 320-detector-row CTP data sets from 9 unilateral MCA AIS cases and 9 controls were retrospectively analyzed. Computed tomography perfusion values for the combined core/penumbra, contralateral hemispheres and arterial territories were measured and compared with literature values. Simple linear regression models are provided to predict corresponding SVD+ value and sSVD CTP values. RESULTS: In the core/penumbra, sSVD generated lower cerebral blood flow (CBF) values, higher mean transit time (MTT) values, and a broader range of CBF and MTT values as compared with SVD+. Mean transit time value differences between the core/penumbra and contralateral hemispheres were statistically significant using sSVD, whereas those of SVD+ were not. Goodness of fit between algorithms for the core/penumbra was lower for CBF (0.483) and MTT (0.494), as compared with time to peak (0.891) and cerebral blood volume (0.997). CONCLUSIONS: In this study using identical source data for patients with MCA AIS, use of either sSVD or SVD+ analyses created statistically significant differences in the CTP value results. Tracer delay-sensitive and -insensitive algorithms impact CTP results in AIS and controls, highlighting the need to pursue additional studies that assess the variability, accuracy, and clinical implications of CTP results generated when using heterogeneous deconvolution algorithms.

Assessment of the tracer delay effect in whole-brain computed tomography perfusion: results in patients without known neuroanatomic abnormalities.

OBJECTIVE: Whole-brain computed tomography perfusion (CTP) data sets generated by tracer delay-insensitive singular value decomposition plus (SVD+) and standard singular value decomposition (sSVD) deconvolution algorithms were evaluated to quantify relatedness and discrepancies in CTP results. METHODS: Twenty females with symmetrical hemispheric CTP maps indicative of brain tissue without apparent abnormalities were studied. Tissue-specific CTP values were analyzed. RESULTS: Standard SVD values were higher than SVD+ for cerebral blood flow. Other CTP values had minimal differences across brain regions. All simple linear regression models were statistically significant (P < 0.05) except for cerebral blood flow in white matter (P = 0.06). Cerebral blood volume had a good model fit, and mean transit time, a poor fit. CONCLUSIONS: Corresponding fitted CTP values for sSVD and SVD+ based on regression equations for brain-tissue types are presented. Additional research is required to compare SVD+ and sSVD in disease states when significant hemodynamic brain alterations are present.

Globalization of P4 medicine: predictive, personalized, preemptive, and participatory--summary of the proceedings of the Eighth International Symposium of the International Society for Strategic Studies in Radiology, August 27-29, 2009.

In August 2009, the International Society for Strategic Studies in Radiology held its eighth biennial meeting. The program focused on the globalization of predictive medicine--or P4 medicine--as it relates to the practice of radiology and radiology research. P4 medicine refers to predictive, personalized, preemptive, and participatory medicine and was the inspiration of Elias Zerhouni, MD, former director of the National Institutes of Health. This article is a summary of some of the key concepts presented at the meeting by an international group of radiologists, imaging scientists, and leaders of industry. In predictive medicine, imaging and imaging-related technologies will likely play an increasing role in the early detection of disease and, thus, the preemption of the development of advanced, hard-to-treat disease. Research into systems biology and molecular imaging promises to personalize medicine, facilitating the provision of the right care to the right patient at the right time. In participatory medicine, increasing interactions with referring physicians and patients will be helpful in raising awareness and recognition of the role of radiologists and will have a positive effect on professionalism. There is also a need to increase awareness of the vital role of radiologists as imaging and radiation safety experts who evaluate the necessity and appropriateness of examinations, monitor performance quality, and are available for postexamination consultations.

A quantitative approach to sequence and image weighting.

Weighting is the term most frequently used to describe magnetic resonance pulse sequences and the concept most commonly used to relate image contrast to differences in magnetic resonance tissue properties. It is generally used in a qualitative sense with the single tissue property thought to be most responsible for the contrast used to describe the weighting of the image as a whole. This article describes a quantitative approach for understanding the weighting of sequences and images, using filters and partial derivatives of signal with respect to logarithms of tissue property values. Univariate and multivariate models are described for several pulse sequences including methods for maximizing weighting and calculating both sequence and image weighting ratios. The approach provides insights into difficulties associated with qualitative use of the concept of weighting and a quantitative basis for assessing the signal, contrast, and weighting of commonly used sequences and images.

Advancing radiology through informed leadership: summary of the proceedings of the Seventh Biannual Symposium of the International Society for Strategic Studies in Radiology (IS(3)R), 23-25 August 2007.

The International Society for Strategic Studies in Radiology (IS(3)R) brings together thought leaders from academia and industry from around the world to share ideas, points of view and new knowledge. This article summarizes the main concepts presented at the 2007 IS(3)R symposium, providing a window onto trends shaping the future of radiology. Topics addressed include new opportunities and challenges in the field of interventional radiology; emerging techniques for evaluating and improving quality and safety in radiology; and factors impeding progress in molecular imaging and nanotechnology and possible ways to overcome them. Regulatory hurdles to technical innovation and drug development are also discussed more broadly, along with proposals for addressing regulators' concerns and streamlining the regulatory process.

Magnetic Resonance Imaging of Normal Pressure Hydrocephalus.

Normal pressure hydrocephalus (NPH) is a syndrome found in the elderly, which is characterized by ventriculomegaly and deep white matter ischemia (DWMI) on magnetic resonance imaging (MRI) and the clinical triad of gait disturbance, dementia, and urinary incontinence. NPH has been estimated to account for up to 10% of cases of dementia and is significant because it is treatable by ventriculoperitoneal shunting. Patients with a known cause of chronic communicating hydrocephalus, that is, meningitis or hemorrhage, tend to respond better than patients with the so-called "idiopathic" form, most likely because of poor selection criteria in the past. Good response to shunting has been associated with hyperdynamic cerebrospinal fluid (CSF) flow through the aqueduct. In the early days of MRI, patients with a large CSF flow void extending from the foramen of Monro through the aqueduct to the fourth ventricle had an excellent chance of responding to ventriculoperitoneal shunting (P < 0.003). Today, we use phase-contrast MRI to measure the volume of CSF flowing through the aqueduct in either direction over a cardiac cycle. When this aqueductal CSF stroke volume is sufficiently elevated, there is an excellent chance of shunt responsiveness (100% positive predictive value in 1 study). Idiopathic NPH appears to be a "two-hit" disease-benign external hydrocephalus (BEH) in infancy followed by DWMI in late adulthood. As BEH occurs when the sutures are still open, these infants present with large heads, a finding also noted in patients with NPH. Although BEH has been attributed to immature arachnoidal granulations with decreased CSF resorptive capacity, this now appears to be permanent and may lead to a parallel pathway for CSF resorption via the extracellular space of the brain. With DWMI, the myelin lipid is lost, exposing the polar water molecules to myelin protein, increasing resistance to CSF outflow and leading to backing up of CSF and hydrocephalus.

Cerebrospinal fluid flow in adults.

This chapter uses magnetic resonance imaging phase-contrast cerebrospinal fluid (CSF) flow measurements to predict which clinical normal-pressure hydrocephalus (NPH) patients will respond to shunting as well as which patients with Chiari I are likely to develop symptoms of syringomyelia. Symptomatic NPH patients with CSF flow (measured as the aqueductal CSF stroke volume) which is shown to be hyperdynamic (defined as twice normal) are quite likely to respond to ventriculoperitoneal shunting. The hyperdynamic CSF flow results from normal systolic brain expansion compressing the enlarged ventricles. When atrophy occurs, there is less brain expansion, decreased aqueductal CSF flow, and less likelihood of responding to shunting. It appears that NPH is a "two-hit" disease, starting as benign external hydrocephalus in infancy, followed by deep white-matter ischemia in late adulthood, which causes increased resistance to CSF outflow through the extracellular space of the brain. Using computational flow dynamics (CFD), CSF flow can be modeled at the foramen magnum and in the upper cervical spine. As in the case of NPH, hyperdynamic CSF flow appears to cause the signs and symptoms in Chiari I and can provide an additional indication for surgical decompression. CFD can also predict CSF pressures over the cardiac cycle. It has been hypothesized that elevated pressure pulses may be a significant etiologic factor in some cases of syringomyelia.

The ependymal "Dot-Dash" sign: an MR imaging finding of early multiple sclerosis.

BACKGROUND AND PURPOSE: Corpus callosum lesions are of specific interest in the evaluation of suspected multiple sclerosis in brain MR imaging. Using thin-section sagittal fluid-attenuated inversion recovery images, researchers have shown that the finding of "subcallosal striations" correlates significantly with the diagnosis of multiple sclerosis. Using the same MR imaging technique, we describe a finding of ependymal irregularity that we call the "Dot-Dash" sign, which we believe to be associated with early multiple sclerosis. METHODS: Sagittal 2-mm fast fluid-attenuated inversion recovery images were obtained in 70 patients. Thirty-five patients had multiple sclerosis according to the Poser criteria, and 35 were age-matched controls. The images were reviewed in a blinded fashion by an experienced neuroradiologist for the presence or absence of the Dot-Dash sign. RESULTS: The correlation between the Dot-Dash sign and definite clinical multiple sclerosis is highly significant (P < .001), with a sensitivity of 91.4% and a specificity of 65.7%. In the age group of < or =50 years, the sensitivity was 95.7% and the specificity, 71.9%. CONCLUSION: The Dot-Dash sign of ependymal irregularity on thin-section sagittal fluid-attenuated inversion recovery images is an early marker for multiple sclerosis, which is particularly useful in the younger patient. This finding appears to be more sensitive for early lesion detection than any other multiple sclerosis imaging finding yet described in the literature.

Image Interpretation Session: Sunday, November 27, 2005.

The Sunday afternoon Image Interpretation Session has been a high point of the annual meeting of the Radiological Society of North America for over 65 years. A panel of five experts has been selected, representing the very best from the fields of neurologic, abdominal, thoracic, pediatric, and musculoskeletal radiology. Each panelist will dazzle us with an insightful analysis of two difficult cases in their area of expertise. The panelists are to be lauded for their bravery in subjecting their diagnostic acumen to the scrutiny of the thousands of radiologists in the audience. The cases, representing a diverse spectrum of diseases and disease manifestations, were selected from recent clinical imaging studies performed at the Stanford University Medical Center or the Lucille Salter Packard Children's Hospital. This session celebrates the skills of diagnostic radiologists worldwide, who are called on daily to amalgamate disparate clinical information with complex imaging data into focused differential diagnoses and effective treatment planning. We hope that these cases will serve to illustrate the central role that expert image interpretation plays in the care of patients. We welcome our audience of RSNA attendees, readers of RadioGraphics, and cyberspace denizens to join with our experts in solving these medical puzzles and to enjoy the excitement of unraveling the unknown.

Magnetic Resonance Imaging and Computed Tomography of the Brain-50 Years of Innovation, With a Focus on the Future.

This review focuses specifically on the developments in brain imaging, as opposed to the spine, and specifically conventional, clinical, cross-sectional imaging, looking primarily at advances in magnetic resonance imaging (MRI) and computed tomography (CT). These fields are viewed from a perspective of landmark publications in the last 50 years and subsequently more in depth using sentinel publications from the last 5 years. It is also written from a personal perspective, with the authors having witnessed the evolution of both fields from their initial clinical introduction to their current state. Both CT and MRI have made tremendous advances during this time, regarding not only sensitivity and spatial resolution, but also in terms of the speed of image acquisition. Advances in CT in recent years have focused in part on reduced radiation dose, an important topic for the years to come. Magnetic resonance imaging has seen the development of a plethora of scan techniques, with marked superiority to CT in terms of tissue contrast due to the many parameters that can be assessed, and their intrinsic sensitivity. Future advances in MRI for clinical practice will likely focus both on new acquisition techniques that offer advances in speed and resolution, for example, simultaneous multislice imaging and data sparsity, and on standardization and further automation of image acquisition and analysis. Functional imaging techniques including specifically perfusion and functional magnetic resonance imaging will be further integrated into the workflow to provide pathophysiologic information that influence differential diagnosis, assist treatment decision and planning, and identify and follow treatment-related changes.

Prostate diffusion imaging with distortion correction.

PURPOSE: Diffusion imaging in the prostate is susceptible to distortion from B0 inhomogeneity. Distortion correction in prostate imaging is not routinely performed, resulting in diffusion images without accurate localization of tumors. We performed and evaluated distortion correction for diffusion imaging in the prostate. MATERIALS AND METHODS: 28 patients underwent pre-operative MRI (T2, Gadolinium perfusion, diffusion at b=800 s/mm(2)). The restriction spectrum protocol parameters included b-values of 0, 800, 1500, and 4000 s/mm(2) in 30 directions for each nonzero b-value. To correct for distortion, forward and reverse trajectories were collected at b=0 s/mm(2). Distortion maps were generated to reflect the offset of the collected data versus the corrected data. Whole-mount histology was available for correlation. RESULTS: Across the 27 patients evaluated (excluding one patient due to data collection error), the average root mean square distortion distance of the prostate was 3.1 mm (standard deviation, 2.2mm; and maximum distortion, 12 mm). CONCLUSION: Improved localization of prostate cancer by MRI will allow better surgical planning, targeted biopsies and image-guided treatment therapies. Distortion distances of up to 12 mm due to standard diffusion imaging may grossly misdirect treatment decisions. Distortion correction for diffusion imaging in the prostate improves tumor localization.

MRI-Derived Restriction Spectrum Imaging Cellularity Index is Associated with High Grade Prostate Cancer on Radical Prostatectomy Specimens.

PURPOSE: We evaluate a novel magnetic resonance imaging (MRI) technique to improve detection of aggressive prostate cancer (PCa). MATERIALS AND METHODS: We performed a retrospective analysis of pre-surgical prostate MRI scans using an advanced diffusion-weighted imaging technique called restriction spectrum imaging (RSI), which can be presented as a normalized z-score statistic. Scans were acquired prior to radical prostatectomy. Prostatectomy specimens were processed using whole-mount sectioning and regions of interest (ROIs) were drawn around individual PCa tumors. Corresponding ROIs were drawn on the MRI imaging and paired with ROIs in regions with no pathology. RSI z-score and conventional apparent diffusion coefficient (ADC) values were recorded for each ROI. Paired t-test, ANOVA, and logistic regression analyses were performed. RESULTS: We evaluated 28 patients with 64 ROIs (28 benign and 36 PCa). The mean difference in RSI z-score (PCa ROI-Benign ROI) was 2.17 (SE = 0.11; p < 0.001) and in ADC was 551 mm(2)/s (SE = 80 mm(2)/s; paired t-test, p < 0.001). The differences in the means among all groups (benign, primary Gleason 3, and primary Gleason 4) was significant for both RSI z-score (F 3,64 = 97.7, p < 0.001) and ADC (F 3,64 = 13.9, p < 0.001). A t-test was performed on only PCa tumor ROIs (n = 36) to determine PCa aggressiveness (Gleason 3 vs. Gleason 4) revealing that RSI z-score was still significant (p = 0.03), whereas, ADC values were no longer significant (p = 0.08). In multivariable analysis adjusting for age and race, RSI z-score was associated with PCa aggressiveness (OR 10.3, 95% CI: 1.4-78.0, p = 0.02) while ADC trended to significance (p = 0.07). CONCLUSION: The RSI-derived normalized cellularity index is associated with aggressive PCa as determined by pathologic Gleason scores. Further utilization of RSI techniques may serve to enhance standardized reporting systems for PCa in the future.

Increased intracranial volume: a clue to the etiology of idiopathic normal-pressure hydrocephalus?

BACKGROUND AND PURPOSE: The etiology of idiopathic normal-pressure hydrocephalus (NPH) is unknown. The purpose of this study was to examine the hypothesis that NPH begins in infancy as benign external hydrocephalus due to decreased uptake of CSF by the arachnoid villi. Since this occurs before the sutures fuse, a secondary hypothesis is that the intracranial volumes of patients with NPH should be larger than those of healthy individuals. METHODS: Intracranial volumes of 51 patients with clinically suspected NPH were compared with those of age- and sex-matched control subjects. All patients underwent phase-contrast CSF velocity MR imaging. They had aqueductal CSF stroke volumes of at least 60 microL, which was 50% higher than previously published normal values. Intracranial volumes were measured and compared between groups. RESULTS: The average intracranial volume for men with NPH (n = 22) was 1682 mL compared with 1565 for male control subjects (n = 55). The NPH volume averaged 118 mL (7.5%) larger than the control volume (P = .003). The average intracranial volume for women with NPH (n = 29) was 1493 mL compared with 1405 mL for female control subjects (n = 55). The NPH volume was 88 mL (6.3%) larger than the control volume (P = .002). CONCLUSION: Patients with NPH have intracranial volumes significantly larger than normal, suggesting that the initial insult occurs before the sutures fuse at 1 year of age. The patients somehow remain asymptomatic until their later years, when a second insult must occur, leading to symptomatic NPH.

Diffusion-weighted imaging in cancer: physical foundations and applications of restriction spectrum imaging.

Diffusion-weighted imaging (DWI) has been at the forefront of cancer imaging since the early 2000s. Before its application in clinical oncology, this powerful technique had already achieved widespread recognition due to its utility in the diagnosis of cerebral infarction. Following this initial success, the ability of DWI to detect inherent tissue contrast began to be exploited in the field of oncology. Although the initial oncologic applications for tumor detection and characterization, assessing treatment response, and predicting survival were primarily in the field of neurooncology, the scope of DWI has since broadened to include oncologic imaging of the prostate gland, breast, and liver. Despite its growing success and application, misconceptions about the underlying physical basis of the DWI signal exist among researchers and clinicians alike. In this review, we provide a detailed explanation of the biophysical basis of diffusion contrast, emphasizing the difference between hindered and restricted diffusion, and elucidating how diffusion parameters in tissue are derived from the measurements via the diffusion model. We describe one advanced DWI modeling technique, called restriction spectrum imaging (RSI). This technique offers a more direct in vivo measure of tumor cells, due to its ability to distinguish separable pools of water within tissue based on their intrinsic diffusion characteristics. Using RSI as an example, we then highlight the ability of advanced DWI techniques to address key clinical challenges in neurooncology, including improved tumor conspicuity, distinguishing actual response to therapy from pseudoresponse, and delineation of white matter tracts in regions of peritumoral edema. We also discuss how RSI, combined with new methods for correction of spatial distortions inherent in diffusion MRI scans, may enable more precise spatial targeting of lesions, with implications for radiation oncology and surgical planning. See all articles in this Cancer Research section, "Physics in Cancer Research."