Machine Learning: Using Xception, a Deep Convolutional Neural Network Architecture, to Implement Pectus Excavatum Diagnostic Tool from Frontal-View Chest X-rays.

Pectus excavatum (PE), a chest-wall deformity that can compromise cardiopulmonary function, cannot be detected by a radiologist through frontal chest radiography without a lateral view or chest computed tomography. This study aims to train a convolutional neural network (CNN), a deep learning architecture with powerful image processing ability, for PE screening through frontal chest radiography, which is the most common imaging test in current hospital practice. Posteroanterior-view chest images of PE and normal patients were collected from our hospital to build the database. Among them, 80% were used as the training set used to train the established CNN algorithm, Xception, whereas the remaining 20% were a test set for model performance evaluation. The performance of our diagnostic artificial intelligence model ranged between 0.976-1 under the receiver operating characteristic curve. The test accuracy of the model reached 0.989, and the sensitivity and specificity were 96.66 and 96.64, respectively. Our study is the first to prove that a CNN can be trained as a diagnostic tool for PE using frontal chest X-rays, which is not possible by the human eye. It offers a convenient way to screen potential candidates for the surgical repair of PE, primarily using available image examinations.

A retrospective study on the impact of bar flipping on the recurrence of pectus excavatum after the Nuss procedure.

BACKGROUND: The Nuss procedure is widely used to correct pectus excavatum. Bar displacement is a common complication associated with this procedure. How the flipping of the bar affects pectus excavatum recurrence has not been reported. In our study, we discuss this and also offer an easier method to determine bar flipping. METHODS: This retrospective study analyzed pectus excavatum patients who underwent primary Nuss repair from August 2014 to December 2018. The preoperative and postoperative Haller indices were measured on chest radiographs (cxrHI). The slope angle of bar flipping (α) was measured on lateral chest radiographs. The improvement index after surgical repair was calculated by: ([preoperative cxrHI-postoperative cxrHI]/preoperative cxrHI × 100). The impact of α on the improvement index was analyzed using one-way analysis of variance and receiver operating characteristic tests. RESULTS: In this study, 359 adult and adolescent patients with an average age of 23.9 ± 7.7 years were included. We formed four subgroups based on the α value: α ≤ 10° (n = 131), α = 11-20° (n = 154), α = 21-30° (n = 51), and α > 30° (n = 23). The mean improvement indices in these groups were 27%, 28%, 26%, and 13%, respectively. Patients with α > 30° were associated with a significantly poorer improvement index than those from the other subgroups (p < 0.001). CONCLUSIONS: The α value is an alternative measurement method for presenting the radiological outcomes after the Nuss procedure. An α > 30° indicates a possible recurrence of pectus excavatum after the Nuss repair. Surgical revision may be considered in patients with an α > 30°, while monitoring should be considered in the other patient groups.

Improving segmentation accuracy of CT kidney cancer images using adaptive active contour model.

In the present study, we retrospectively analyzed the records of surgical confirmed kidney cancer with renal cell carcinoma pathology in the database of the hospital. We evaluated the significance of cancer size by assessing the outcomes of proposed adaptive active contour model (ACM). The aim of our study was to develop an adaptive ACM method to measure the radiological size of kidney cancer on computed tomography in the hospital patients. This paper proposed a set of medical image processing, applying images provided by the hospital and select the more obvious cases by the doctors, after the first treatment to remove noise image, and the kidney cancer contour would be circled by using the proposed adaptive ACM method. The results showed that the experimental outcome has highly similarity with the medical professional manual contour. The accuracy rate is higher than 99%. We have developed a novel adaptive ACM approach that well combines a knowledge-based system to contour the kidney cancer size in computed tomography imaging to support the clinical decision.

Differential diagnosis of atypical encephalopathy in critical care: a case report.

BACKGROUND: A lower level of consciousness is a common presentation in critical care, with many different causes and contributory factors, of which more than one may be present concurrently. CASE PRESENTATION: We described a woman with poorly controlled diabetes and steroid-dependent asthma who presented in a deep coma. She was found to have Streptococcus intermedius bacteremia and pyogenic ventriculitis that originated from right middle lobe pneumonia. Also, multiple small parenchymal lesions were observed on brain magnetic resonance imaging and increased protein concentration was noted in cerebral spinal fluid. Initially, her coma was thought to be due to diabetic ketoacidosis and septic encephalopathy. However, her lowered level of consciousness was disproportionate to either diabetic ketoacidosis or septic encephalopathy, and her clinical course was not as expected for these two conditions. Treatment with antibiotic, corticosteroid and antihelminthic drugs was administered resulting in improving consciousness. The Streptococcus intermedius pneumonia progressed to form a large cavity that needed an early surgical lobectomy and resulted in the unexpected diagnosis of chronic cavitary pulmonary aspergillosus. CONCLUSIONS: In critical care, a lowered level of consciousness may have many etiologies, and critical care clinicians should be familiar with the signs and symptoms of all possible causes to enable prompt diagnosis and appropriate treatment.

The neural generators of the mismatch responses to Mandarin lexical tones: an MEG study.

The present magnetoencephalography study used the cortically constrained minimum-norm estimates of human brain activity to elucidate functional roles of neural generators for detecting different magnitudes of lexical tones changes. A multiple-deviant oddball paradigm was used in which the syllable "yi" with a low-dipping tone (T3) was the common standard sound and the same syllable with a high-level tone (T1) or a high-rising tone (T2) were the large and small deviant sounds, respectively. The data revealed a larger magnetic mismatch field (MMNm) for large deviant in the left hemisphere. The source analysis also confirmed that the MMNm to lexical tone changes was generated in bilateral superior temporal gyri and only the large deviant revealed left lateralization. A set of frontal generators was activated at a later time and revealed differential sensitivities to the degree of deviance. The left anterior insula, the right anterior cingulate cortex, and the right ventral orbital frontal cortex were activated when detecting a large deviant, whereas the right frontal-opercular region was sensitive to the small deviant. These frontal generators were thought to be associated with various top-down mechanisms for attentional modulation. The time frequency (TF) analysis showed that large deviants yielded large theta band (5-7Hz) activity over the left anterior scalp and the left central scalp, while small deviants yielded large alpha band activity (9-11Hz) over the posterior scalp. The results of TF analyses implied that mechanisms of working memory and functional inhibition involved in the processes of acoustic change detection.

Surgical Treatment of Castleman Disease Using Cardiopulmonary Bypass.

UNLABELLED: Castleman disease is a rare benign lymphoproliferative disorder of unknown etiology. A 29-year-old woman presented with intermittent chest pain for 2 years. Upon examination, computed tomography showed an intensely enhanced solid mass that encased her right pulmonary artery. The tumor was resected safely and completely via standard thoracotomy with cardiopulmonary bypass. KEY WORDS: Castleman disease; Middle mediastinal tumor; Surgery.

Resting-state functional magnetic resonance imaging analysis with seed definition constrained by regional homogeneity.

Researchers have recently focused their attention on the intrinsic functional connectivity (FC) in the brain using resting-state functional magnetic resonance imaging. Seed-based correlation analysis (SCAC), which correlates a predefined seed region with other voxels in the brain, is a common index for FC. However, definition of seed sizes and locations was ambiguous in previous studies and this may lead to spurious results for people with a unique functional anatomy. To address this issue, this study proposes a novel method (SCAReHo) that provides a data-driven seed selection (including sizes and locations) method by incorporating regional homogeneity (ReHo) in the SCAC method. The disparities between SCAC and SCAReHo methods among 12 healthy participants were evaluated in the FC of default mode network (DMN), task-positive network (TPN), and amygdala network. The SCAReHo method bypasses the seed-selection ambiguity and enhances the sensitivity in detecting FC of the DMN, TPN, and amygdala network. This study suggests that the SCAReHo method improves the sensitivity of FC analysis and reduces the uncertainty of seed selection. Thus, this method may be particularly useful for psychiatric and neurological investigations.

Vascular space occupancy MRI during breathholding at 3 Tesla.

PURPOSE: To evaluate the vasodilatory response of normal human brain and meningiomas under repeated breathholding challenges using vascular space occupancy (VASO) MRI at 3 Tesla (T). MATERIALS AND METHODS: Five normal volunteers and five patients with meningiomas were recruited for this study. For the normal group, VASO MRI during repeated breathholds of different duration (5 to 30 s) was acquired. Patients performed a 15-s breathhold paradigm for VASO MRI. The maximum signal change and full-width at half-maximum (FWHM) were determined by curve fitting. RESULTS: Significant VASO signal decreases in the gray matter could be detected for a breathhold period as short as 5 s. The fractional activation volume vs. breathhold duration reached a plateau around 34.21 ± 3.39% at 15 s. In the patient group, there were significant VASO signal decreases in normal gray matters and also in small areas of three large-sized meningiomas. CONCLUSION: The 3T VASO MRI detected significant signal decreases in the gray matter, but not in the white matter, during short periods of breathholding. The fractional activation volume reached the plateau at 15-s breathhold, which is recommended for clinical application.

Is Weisskoff model valid for the correction of contrast agent extravasation with combined T1 and T2* effects in dynamic susceptibility contrast MRI?

PURPOSE: The Weisskoff model has been widely applied for correcting the T1 effect of the contrast agent leakage in the measured dynamic susceptibility contrast (DSC)-MRI signals. This study aimed to modify the Weisskoff model for the inclusion of both T1 and T2 effects of the contrast agent extravasation. METHODS: A two-compartment model was proposed and implemented into the original Weisskoff model to describe the combined T1 and T2 effects from the contrast agent leakage in the measured DSC-MRI signals. A computer simulation was performed to evaluate the dependence of T, versus T2 dominance on imaging parameter, field strength, baseline T1, and severity of the leakage. The modified Weisskoff model was employed to correct the relative cerebral blood volume (rCBV) maps in three patients with brain tumors to demonstrate its use. RESULTS: The resultant equation had the same mathematical form as the original model, but with a different expression for the fitting constant K2. This new parameter can be of either a positive or a negative value. Results of the computer simulation showed more probable T2 dominance with longer TE, higher field strength, shorter baseline T1, and greater extraction of the contrast agent. Clinical data were well fitted by the model, with a positive K2 indicating T1 dominance and underestimated rCBV and a negative K2 indicating T2 dominance and overestimated rCBV. The K2 values of normal-appearing brain tissues were distributed in a much smaller range than the K2 values of enhancing tumors. The ratios of corrected over uncorrected normalized CBV (nCBV) for gray matter (GM) were in the range between 1.04 and 1.05, meaning that the nCBV remained rather stable before and after correction. The ratios for the tumors were 0.65, 0.42, and 2.81, either much smaller or greater than the ratios for GM. CONCLUSIONS: This study proposed a modified Weisskoff model that was able to explain both T1 and T2 dominant effects of the contrast agent extravasation in DSC-MRI. Further development is needed to make the K2 parameter a quantitative indicator of the vessel permeability.

A subtle grey-matter increase in first-episode, drug-naive major depressive disorder with panic disorder after 6 weeks' duloxetine therapy.

We designed this study to investigate the modulating effects of duloxetine on symptoms and grey matter of patients with major depressive disorder combined with panic disorder. We also aimed to discover if there was any persistence of grey-matter deficits after remission and to find 'trait markers' for this comorbidity. High-resolution magnetic resonance imaging and voxel-based morphometric measurements were performed on 15 patients at baseline and remitted status (week 6) compared to 15 healthy control subjects who were scanned twice within 6 wk. The rating scales of depressive and panic symptoms improved with statistical significance (corrected p<0.001). A widespread pattern of grey-matter deficits in infero-frontal, limbic, occipital, temporo-parietal, cerebellar areas (trait marker regions) in drug-naive patients were observed compared to controls at baseline (family-wise error corrected p<0.0002). There were no significant changes of grey matter in healthy controls over the 6-wk period. Duloxetine-induced increases of grey matter were very subtle in left infero-frontal cortex, right fusiform gyrus, and right cerebellum VIIIa areas (state marker regions) after 6-wk therapy (uncorrected p<0.0005). Duloxetine did not increase grey matter to the level of control subjects and grey-matter deficits in patients appear largely unaffected by duloxetine. We suggest that short-term duloxetine therapy improved the clinical symptoms of patients with major depressive disorder combined with panic disorder. These improvements might be related to a modest increase of grey matter in state marker regions of the brain. The deficits of trait marker regions were more evident and are likely to be important for pathogenesis.

First episode drug-naïve major depressive disorder with panic disorder: gray matter deficits in limbic and default network structures.

This study was designed to investigate the structural differences in the brains of first episode, drug-naïve patients with major depressive disorder and panic disorder versus healthy control subjects. High-resolution brain magnetic resonance images were performed on patients and health control subjects (age, sex and handedness matched). Structural magnetic resonance images of brain were estimated by optimized voxel-based morphometry of FSL (FMRIB Software Library). Patients had deficits of gray matter volumes over right anterior cingulate cortex, right medial frontal gyrus, left posterior cingulate cortex, right parahippocampal gyrus, limbic areas, occipital lingual gyrus and bilateral cerebellums when compared to controls. These results suggested that this group of patients has possible deficits of gray matter volumes over the default-mode network, fronto-cingulate and limbic structures. The decline of gray matter volumes might have started since the first episode.

Study-specific EPI template improves group analysis in functional MRI of young and older adults.

Spatial normalization to a common coordinate space, e.g. via the Montreal Neurological Institute (MNI) brain template, is an essential step of analyzing multi-subject functional MRI (fMRI) datasets. The imperfect compensation for individual regional discrepancies during spatial transformation, which could potentially introduce localization errors of the activation foci and/or reduce the detection sensitivity, may be minimized if a template specifically designed for the subjects of a study is applied. In this fMRI study, we proposed and evaluated the use of a study-specific template (SST) based on the mean of individually normalized echo-planar images for group data analysis. A hand flexion and a word generation tasks were performed on young volunteers in experiment 1. Comparing with the MNI template approach, greater t-values of local maxima and activated voxels were detected within volume-of-interests (VOIs) with the SST approach in both tasks. Moreover, the SST approach reduced Euclidean distances between activation foci of individuals and group by 1.52 mm in motor fMRI and 5.84 mm in language fMRI. Similar results were obtained with or without spatial smoothing of the echo-planar images. Experiment 2 further examined these two approaches in older adults, in which volumetric differences between subjects are of great concerns. With a working memory task, the SST approach showed greater t-values of local maxima and activated voxels within the VOI of prefrontal gyrus. This study demonstrated that the SST resulted in more focused activation patterns and effectively improved the fMRI sensitivity, which suggested potentials of reducing number of subjects required for group analysis.

Breathhold-regulated blood oxygenation level-dependent (BOLD) MRI of human brain at 3 tesla.

PURPOSE: To investigate the cerebrovascular response to repeated breathhold challenges using blood oxygenation level-dependent (BOLD) MRI at 3T and compare the results with previous data at 1.5T. MATERIALS AND METHODS: Six normal volunteers and six patients with brain tumors were recruited for this 3T study. For the normal group, BOLD MRI during repeated breathholds of different durations (five to 30 seconds) were acquired. Maximum signal change, full-width at half-maximum (FWHM) and onset time (defined as the time to the first half-maximum) were determined by curve fitting. The fractional activation volume was also calculated. Patients performed a 10- or 15-second breathhold paradigm according to individual capability. RESULTS: Significant BOLD signal increases in the gray matter for a breathhold period as short as 5 seconds at 3T, instead of 10 seconds at 1.5T. The fractional activation volume vs. breathhold duration reached a plateau of 49.54 +/- 7.26% at 15 seconds at 3T, which was higher and shorter than that at 1.5T. The maximum signal changes were significantly larger (a 69% increase) at 3T than at 1.5T. In the patient group, there were BOLD signal increases in gray matter but not in tumor bulk or perifocal edema, which agreed with the results previously found at 1.5T. CONCLUSION: BOLD MRI at 3T is more sensitive for detecting breathhold-regulated signal changes than at 1.5T, which allows a shorter and more feasible breathhold paradigm for clinical applications in patients with brain tumors.

Vessel size imaging using dual contrast agent injections.

PURPOSE: To investigate the feasibility of a vessel size imaging (VSI) technique with separate contrast agent injections for evaluation of the vessel caliber in normal tissues and in brain tumors. MATERIALS AND METHODS: Computer simulation was first performed to assess the potential errors in the estimation of vessel caliber that could result from time shifts between the dual contrast agent injections. Eight patients (four female, four male, 37-77 years old) with brain tumors (three high-grade gliomas, two low-grade gliomas, and three meningiomas) were recruited for clinical study. Dynamic susceptibility contrast magnetic resonance imaging (MRI) using gradient echo (GE) and spin echo (SE) echo-planar imaging sequences were performed separately with a 10-minute interval on a 3.0T scanner. Vessel caliber maps were calculated and analyzed in regions of interest at cortical gray matter (GM), thalamus, white matter (WM), and tumors. RESULTS: From the computer simulation, the error of vessel caliber measurement was less than 8% when the difference between the time-to-peak of the GE and the SE studies was 1.5 seconds, and reduced to within 5% when the difference was 1 second. From the patient datasets of a 64 x 64 matrix, the estimated vessel calibers were 37.4 +/- 12.9 microm for cortical gray matter, 20.7 +/- 8.8 microm for thalamus, and 15.0 +/- 5.1 microm for white matter, comparable to results in the literature. Two patients had a VSI with 128 x 128 matrix and showed similar results in vessel calibers of normal tissues. All the tumors had larger mean vessel diameter than normal-appearing tissues. The difference in vascular size between normal tissue and tumor was demonstrated clearly in both the VSIs of regular and high spatial resolution. CONCLUSION: This study suggests that VSI with a dual injection method is a feasible technique for estimating microvascular calibers of normal tissues and brain tumors in clinical scanners.

False-positive analysis of functional MRI during simulated deep brain stimulation: a phantom study.

PURPOSE: To investigate the false-positive activations/deactivations in functional MRI (fMRI) of deep brain stimulation (DBS) using a phantom. MATERIALS AND METHODS: fMRI experiments were performed on a 1.5T scanner using a single-shot gradient-echo echo-planar imaging (GE-EPI) sequence (TR/TE/FA = 6000 msec/60 msec/90 degrees ) on an agar-gel phantom inserted with DBS electrodes. During the experimental blocks, two-second stimuli were delivered during the interscan waiting time (ISWT), which was adjusted by changing the number of slices acquired within the TR (3500 msec with 30 slices and 5160 msec with 10 slices). Data were analyzed using SPM2 software, and the false-positive voxels were detected with five different P-value thresholds. RESULTS: The number of false-positive voxels in experimental conditions had no significant differences from those in control conditions with either long or short ISWT, which increased with the P-value threshold from zero at P < 0.0001 to approximately 40 at P < 0.05. The pattern of increasing number of false-positive reactions along with P-value was similar between all conditions. CONCLUSION: False-positive findings from fMRI with similar experimental design can be well controlled with a statistical threshold of P < 0.001 or tighter. The short ISWT of 3500 msec did not increase false-positive reactions compared to the long ISWT of 5160 msec.

Expertise modulates the perception of pain in others.

Perceiving the pain of others activates a large part of the pain matrix in the observer [1]. Because this shared neural representation can lead to empathy or personal distress [2, 3], regulatory mechanisms must operate in people who inflict painful procedures in their practice with patient populations in order to prevent their distress from impairing their ability to be of assistance. In this functional magnetic resonance imaging MRI study, physicians who practice acupuncture were compared to naive participants while observing animated visual stimuli depicting needles being inserted into different body parts, including the mouth region, hands, and feet. Results indicate that the anterior insula somatosensory cortex, periaqueducal gray, and anterior cingulate cortex were significantly activated in the control group, but not in the expert group, who instead showed activation of the medial and superior prefrontal cortices and the temporoparietal junction, involved in emotion regulation and theory of mind.

A man in his mid-70s with a sellar mass.

Metastatic melanoma to a pituitary oncocytoma is a very rare condition. A 76-year-old man was presented with progressive visual disturbance and falling down with initial loss of consciousness 2 days before admission. He had a subungual acral lentiginous melanoma (T3N1M0) with gangrenous change of left big toe, treated by amputation 15 months ago. Computed tomography and MR imaging demonstrated masses involving inguina, mediastinum and left renal hilum and dumb-bell shaped hyperdense mass, approximately 6.2 x 3.7 mm, that involved pituitary fossa and suprasellar region with adjacent bony destruction. He underwent surgical resection of the tumor. Microscopically, the tumor revealed an admixture of pituitary adenoma and invasive metastatic melanoma with fragments containing both populations in juxtaposition. The adenoma was negative for melanoma markers and pituitary hormone markers. The melanoma was positive for S-100 protein and BMB-45. Ultrastructure of the adenoma revealed abundant mitochondria and sparse secretory granules. The diagnosis was metastatic melanoma to a pituitary oncocytoma. The current literature on metastatic tumors to pituitary adenoma is reviewed.

Functional MRI of conventional and anomalous metaphors in Mandarin Chinese.

This study looks at whether conventional and anomalous metaphors are processed in different locations in the brain while being read when compared with a literal condition in Mandarin Chinese. We find that conventional metaphors differ from the literal condition with a slight amount of increased activation in the right inferior temporal gyrus. In addition, when the anomalous metaphor condition is compared with the literal condition, increased activation occurs bilaterally in the frontal and temporal gyri. Lastly, the comparison between the anomalous and conventional metaphor conditions shows bilateral activation in the middle frontal gyrus and the precentral gyrus, and right-hemisphere activation in the superior frontal gyrus. Left hemisphere activation is found in the inferior frontal gyrus and fusiform gyrus. The left hemisphere activation in the frontal and temporal gyri point to the recruitment of traditional language-based areas for anomalous metaphor sentences, while the right-hemisphere activation found suggests that remote associations are being formed. In short, our study supports the idea that metaphors are not a homogenous type of figurative language and that distinguishing between different types of metaphors will advance theories of language comprehension.

Diagnosis of malpositioned chest tubes after emergency tube thoracostomy: is computed tomography more accurate than chest radiograph?

OBJECTIVE: The aim of this study was to compare the accuracy between computed tomography (CT) and frontal chest radiography in the diagnosis of malpositioned chest tubes (MCT). MATERIALS AND METHODS: CT scans positive for MCT between March 2000 and March 2004 were reviewed. Two radiologists assessed for intra- and extrathoracic locations of MCT in CT studies. Two physicians who were blinded to the results of CT scans assessed the frontal chest radiographs for location of chest tubes, within the pleural space or outside pleural space. The results of CT were then compared with the results of frontal chest radiographs. Medical records were also reviewed for function of the chest tubes and any complications induced by MCT. RESULTS: CT revealed 28 MCT among the 76 chest tubes that were placed in 54 patients. Among the 28 MCT detected by CT, 23 tubes were in the intrathoracic location (20 intraparenchymal; 3 intrafissural) and 5 tubes were in the extrathoracic location (4 in mediastinum; 1 in chest wall). Frontal chest radiographs only revealed six MCT. Among 28 MCT, 16 sufficient, 8 insufficient, and 4 indeterminate functions of the chest tubes were noted from medical charts. One patient complicated with lung abscess, four patients had suffered pleural empyema, and one patient suffered active lung parenchymal bleeding, resulting from MCT. CONCLUSIONS: CT is more accurate than chest radiograph for the diagnosis of MCT. For selected patients with inadequacy drainage of the tubes and when chest radiograph is noncontributory, CT scan is recommended to clarify the exact location of chest tubes.

Right ventricular dysfunction secondary to acute massive pulmonary embolism detected by helical computed tomography pulmonary angiography.

OBJECTIVE: Acute massive pulmonary embolism causes abrupt pulmonary arterial hypertension and right ventricular dysfunction (RVD). Patients with RVD have a worse prognosis than those with normal right ventricular function. Consequently, recognizing the RVD at the time of pulmonary embolism is useful for risk stratification and enables more aggressive therapy. The study compared the accuracy of helical computed tomographic (CT) scans with echocardiography in the detecting of RVD in patients with acute massive pulmonary embolism. MATERIALS AND METHODS: Specifically, this work reviewed the CT pulmonary angiograms of 14 patients who were positive for acute massive pulmonary embolism during a 52-month period. CT scans were reviewed for findings indicating RVD. Scans were considered positive for RVD if the right ventricle was dilated or there was leftward shift of the interventricular septum. Echocardiographic reports serving as the reference standard for the diagnosis of RVD were also reviewed. CT study results were then correlated with echocardiography results. RESULTS: Among 14 patients with massive pulmonary embolism, echocardiography identified 12 patients having RVD, whereas the remaining two patients were negative for RVD. Meanwhile, CT correctly identified 11 of 12 patients as having RVD, and was negative for RVD in the remaining 3 patients. Correlated with echocardiography, CT scan for RVD detection had a sensitivity of 91.6% and a specificity of 100%. CONCLUSIONS: CT can accurately detect RVD in patients with acute massive pulmonary embolism. However, this result requires confirmation using a larger prospective cohort study.