Multidimensional predicting model of intracranial aneurysm stability with backpropagation neural network: a preliminary study.

OBJECTIVES: The stability of intracranial aneurysms (IAs) may involve in multidimensional factors. Backpropagation (BP) neural network could be adopted to support clinical work. This preliminary study aimed to delve into the feasibility of BP neural network in assessing the risk of IA rupture/growth and to prove the advantage of multidimensional model over single/double-dimensional model. METHODS: Thirty-six IA patients were recruited from a prospective registration study (ChiCTR1900024547). All patients were followed up until aneurysm ruptured/grew or 36 months after being diagnosed with the IAs. The multidimensional data regarding clinical, morphological, and hemodynamic characteristics were acquired. Hemodynamic analyses were conducted with patient-specific models. Based on these characteristics, seven models were built with BP neural network (the ratio of training set to validation set as 8:1). The area under curves (AUC) was calculated for subsequent comparison. RESULTS: Forty-five characteristics were determined from 36 patients with 37 IAs. In the models based on the single dimension of IA characteristics, only morphological characteristics exhibited high performance in assessing 3-year IA stability (AUC = 0.703, P = 0.035). Among the models integrating two dimensions of IA characteristics, clinical-morphological (AUC = 0.731, P = 0.016), clinical-hemodynamic (AUC = 0.702, P = 0.036), and morphological-hemodynamic (AUC = 0.785, P = 0.003) models were capable of assessing the risk of 3-year IA rupture/growth. Moreover, the models including all three dimensions exhibited the maximum predicting significance (AUC = 0.811, P = 0.001). CONCLUSION: The present preliminary study reported that BP neural network might support assessing the 3-year stability of IAs. Models based on multidimensional characteristics could improve the assessment accuracy for IA rupture/growth.

Automatic opportunistic osteoporosis screening using low-dose chest computed tomography scans obtained for lung cancer screening.

OBJECTIVE: Osteoporosis is a prevalent and treatable condition, but it remains underdiagnosed. In this study, a deep learning-based system was developed to automatically measure bone mineral density (BMD) for opportunistic osteoporosis screening using low-dose chest computed tomography (LDCT) scans obtained for lung cancer screening. METHODS: First, a deep learning model was trained and tested with 200 annotated LDCT scans to segment and label all vertebral bodies (VBs). Then, the mean CT numbers of the trabecular area of target VBs were obtained based on the segmentation mask through geometric operations. Finally, a linear function was built to map the trabecular CT numbers of target VBs to their BMDs collected from approved software used for osteoporosis diagnosis. The diagnostic performance of the developed system was evaluated using an independent dataset of 374 LDCT scans with standard BMDs and osteoporosis diagnosis. RESULTS: Our deep learning model achieved a mean Dice coefficient of 86.6% for VB segmentation and 97.5% accuracy for VB labeling. Line regression and Bland-Altman analyses showed good agreement between the predicted BMD and the ground truth, with correlation coefficients of 0.964-0.968 and mean errors of 2.2-4.0 mg/cm3. The area under the curve (AUC) was 0.927 for detecting osteoporosis and 0.942 for distinguishing low BMD. CONCLUSION: The proposed deep learning-based system demonstrated the potential to automatically perform opportunistic osteoporosis screening using LDCT scans obtained for lung cancer screening. KEY POINTS: • Osteoporosis is a prevalent but underdiagnosed condition that can increase the risk of fracture. • A deep learning-based system was developed to fully automate bone mineral density measurement in low-dose chest computed tomography scans. • The developed system achieved high accuracy for automatic opportunistic osteoporosis screening using low-dose chest computed tomography scans obtained for lung cancer screening.

Hematoma volume measurement in gradient echo MRI using quantitative susceptibility mapping.

BACKGROUND AND PURPOSE: A novel quantitative susceptibility mapping (QSM) processing technology has been developed to map tissue susceptibility property without blooming artifacts. We hypothesize that hematoma volume measurement on QSM is independent of imaging parameters, eliminating its echo time dependence on gradient echo MRI. METHODS: Gradient echo MRI of 16 patients with intracerebral hemorrhage was processed with susceptibility-weighted imaging, R2* (=1/T2*) mapping, and QSM at various echo times. Hematoma volumes were measured from these images. RESULTS: Linear regression of hematoma volume versus echo time showed substantial slopes for gradient echo magnitude (0.45±0.31 L/s), susceptibility-weighted imaging (0.52±0.46), and R2* (0.39±0.30) but nearly zero slope for QSM (0.01±0.05). At echo time=20 ms, hematoma volume on QSM was 0.80× that on gradient echo magnitude image (R2=0.99). CONCLUSIONS: QSM can provide reliable measurement of hematoma volume, which can be performed rapidly and accurately using a semiautomated segmentation tool.

Reducing the object orientation dependence of susceptibility effects in gradient echo MRI through quantitative susceptibility mapping.

This study demonstrates the dependence of non-local susceptibility effects on object orientation in gradient echo MRI and the reduction of non-local effects by deconvolution using quantitative susceptibility mapping. Imaging experiments were performed on a 3T MRI system using a spoiled 3D multi-echo GRE sequence on phantoms of known susceptibilities, and on human brains of healthy subjects and patients with intracerebral hemorrhages. Magnetic field measurements were determined from multiple echo phase data. To determine the quantitative susceptibility mapping, these field measurements were deconvolved through a dipole inversion kernel under a constraint of consistency with the magnitude images. Phantom and human data demonstrated that the hypointense region in GRE magnitude image corresponding to a susceptibility source increased in volume with TE and varied with the source orientation. The induced magnetic field extended beyond the susceptibility source and varied with its orientation. In quantitative susceptibility mapping, these blooming artifacts, including their dependence on object orientation, were reduced, and the material susceptibilities were quantified.

[High precision all-reflection Fourier transform imaging spectrometer spectral calibration using homogeneous broadening of the wave number model].

All-reflection Fourier transform imaging spectrometer (ARFTIS) is a novel imaging spectrometer. The specialty is not only high spectrum resolution, but also wide band and non-chromatism. It is good for remote sensing field of wide band imaging. Single spectrum calibration, average calibration and weighted average calibration are three common calibration methods. However, they all are limited. Because they cannot meet the demand on both convenience and high precision. In the present paper, the authors propose a novel model for spectrum calibration. It can work in high precision with single spectrum calibration. At the same time, the method is steady, and the average error is less than 5% with multi-bands calibration. It provides a convenient way for the non-professional calibration situation and outer simply calibration work.

[Spectral reconstruction using algebraic reconstruction technology for Fourier transform computed tomography imaging spectrometer].

In the present paper, the authors will introduce our research on spectral reconstruction of Fourier transform computed tomography imaging spectrometer by means of the algebraic reconstruction technology. A simulation experiment was carried out to demonstrate the algorithm. The spatial similarities and spectral similarities were evaluated using the normalized correlation coefficient. The performance of ART was evaluated when the quantity of projection is 45. In that case, filter back projection can't work well. Actual spectral slices were reconstructed by using ART in the last part of this paper.