In the present article, we provide an overview on the basics of deep learning in terms of technical aspects and steps required to launch a deep learning research. Deep learning is a branch of ...artificial intelligence, which has been attracting interest in many domains. The essence of deep learning can be compared to teaching an elementary school student how to differentiate magnetic resonance images, and we first explain the concept using this analogy. Deep learning models are composed of many layers including input, hidden, and output ones. Convolutional neural networks are suitable for image processing as convolutional and pooling layers allow successfully performing extraction of image features. The process of conducting a research work with deep learning can be divided into the nine following steps: computer preparation, software installation, specifying the function, data collection, data edits, dataset creation, programming, program execution, and verification of results. Concerning widespread expectations, deep learning cannot be applied to solve tasks other than those set in specification; moreover, it requires a large amount of data to train and has difficulties with recognizing unknown concepts. Deep learning cannot be considered as a universal tool, and researchers should have thorough understanding of the features of this technique.
In patients on maintenance dialysis, cardiovascular mortality risk is remarkably high, which can be partly explained by severe coronary artery calcification (CAC). Hyperphosphatemia has been reported ...to be associated with the severity of CAC. However, the optimal phosphate range in patients on dialysis remains unknown. This study was planned to compare the effects on CAC progression of two types of noncalcium-based phosphate binders and of two different phosphate target ranges.
We conducted a randomized, open-label, multicenter, interventional trial with a two by two factorial design. A total of 160 adults on dialysis were enrolled and randomized to the sucroferric oxyhydroxide or lanthanum carbonate group, with the aim of reducing serum phosphate to two target levels (3.5-4.5 mg/dl in the strict group and 5.0-6.0 mg/dl in the standard group). The primary end point was percentage change in CAC scores during the 12-month treatment.
The full analysis set included 115 patients. We observed no significant difference in percentage change in CAC scores between the lanthanum carbonate group and the sucroferric oxyhydroxide group. On the other hand, percentage change in CAC scores in the strict group (median of 8.52; interquartile range, -1.0-23.9) was significantly lower than that in the standard group (median of 21.8; interquartile range, 10.0-36.1;
=0.006). This effect was pronounced in older (aged 65-74 years) versus younger (aged 20-64 years) participants (
value for interaction =0.003). We observed a similar finding for the absolute change in CAC scores.
Further study with a larger sample size is needed, but strict phosphate control shows promise for delaying progression of CAC in patients undergoing maintenance hemodialysis.
Evaluate the New Phosphate Iron-Based Binder Sucroferric Oxyhydroxide in Dialysis Patients with the Goal of Advancing the Practice of EBM (EPISODE), jRCTs051180048.
The objective of our study was to assess the effect of the combination of deep learning-based denoising (DLD) and iterative reconstruction (IR) on image quality and Lung Imaging Reporting and Data ...System (Lung-RADS) evaluation on chest ultra-low-dose CT (ULDCT).
Forty-one patients with 252 nodules were evaluated retrospectively. All patients underwent ULDCT (mean ± SD, 0.19 ± 0.01 mSv) and standard-dose CT (SDCT) (6.46 ± 2.28 mSv). ULDCT images were reconstructed using hybrid iterative reconstruction (HIR) and model-based iterative reconstruction (MBIR), and they were postprocessed using DLD (i.e., HIR-DLD and MBIR-DLD). SDCT images were reconstructed using filtered back projection. Three independent radiologists subjectively evaluated HIR, HIR-DLD, MBIR, and MBIR-DLD images on a 5-point scale in terms of noise, streak artifact, nodule edge, clarity of small vessels, homogeneity of the normal lung parenchyma, and overall image quality. Two radiologists independently evaluated the nodules according to Lung-RADS using HIR, MBIR, HIR-DLD, and MBIR-DLD ULDCT images and SDCT images. The median scores for subjective analysis were analyzed using Wilcoxon signed rank test with Bonferroni correction. Intraobserver agreement for Lung-RADS category between ULDCT and SDCT was evaluated using the weighted kappa coefficient.
In the subjective analysis, ULDCT with DLD showed significantly better scores than did ULDCT without DLD (
< 0.001), and MBIR-DLD showed the best scores among the ULDCT images (
< 0.001) for all items. In the Lung-RADS evaluation, HIR showed fair or moderate agreement (reader 1 and reader 2: κw = 0.46 and 0.32, respectively); MBIR, moderate or good agreement (κw = 0.68 and 0.57); HIR-DLD, moderate agreement (κw = 0.53 and 0.48); and MBIR-DLD, good agreement (κw = 0.70 and 0.72).
DLD improved the image quality of both HIR and MBIR on ULDCT. MBIR-DLD was superior to HIR_DLD for image quality and for Lung-RADS evaluation.
Objectives
To develop a deep learning–based algorithm to detect aortic dissection (AD) and evaluate the diagnostic ability of the algorithm compared with those of radiologists.
Methods
Included in ...the study were 170 patients (85 with AD and 85 without AD). An AD detection algorithm was developed using a convolutional neural network with Xception architecture. Of the patient data, 80% were used for training and validation and 20% were used for testing. Fivefold cross-validation was performed to evaluate the method. An average of 6688 non-contrast-enhanced CT images (slice thickness, 5 mm) were used for training. A radiologist reviewed both contrast-enhanced and non-contrast-enhanced images and identified the slices of AD. The identified slices were used as ground truth. Receiver operating characteristic curve and area under the curve (AUC) analysis was performed. Five radiologists independently evaluated the images. The accuracy, sensitivity, and specificity of the algorithm and those of the radiologists were compared.
Results
The AUC of the developed algorithm was 0.940, and a cutoff value of 0.400 provided accuracy of 90.0%, sensitivity of 91.8%, and specificity of 88.2%. For the radiologists, median (range) accuracy, sensitivity, and specificity were 88.8 (83.5–94.1)%, 90.6 (83.5–94.1)%, and 94.1 (72.9–97.6)%, respectively. There was no significant difference in performance in terms of accuracy, sensitivity, or specificity between the algorithm and the average performance of the radiologists (
p
> 0.05).
Conclusions
The developed algorithm showed comparable diagnostic performance to radiologists for detecting AD, which suggests the potential of the proposed method to support clinical practice by reducing missed ADs.
Key Points
• A deep learning–based algorithm for detecting aortic dissection was developed using the non-contrast-enhanced CT images of 170 patients.
• The algorithm had an AUC of 0.940 for detecting aortic dissection.
• The accuracy, sensitivity, and specificity of the algorithm were comparable to those of radiologists.
Objectives
To compare the image quality of the lungs between ultra-high-resolution CT (U-HRCT) and conventional area detector CT (AD-CT) images.
Methods
Image data of slit phantoms (0.35, 0.30, and ...0.15 mm) and 11 cadaveric human lungs were acquired by both U-HRCT and AD-CT devices. U-HRCT images were obtained with three acquisition modes: normal mode (U-HRCT
N
: 896 channels, 0.5 mm × 80 rows; 512 matrix), super-high-resolution mode (U-HRCT
SHR
: 1792 channels, 0.25 mm × 160 rows; 1024 matrix), and volume mode (U-HRCT
SHR-VOL
: non-helical acquisition with U-HRCT
SHR
). AD-CT images were obtained with the same conditions as U-HRCT
N
. Three independent observers scored normal anatomical structures (vessels and bronchi), abnormal CT findings (faint nodules, solid nodules, ground-glass opacity, consolidation, emphysema, interlobular septal thickening, intralobular reticular opacities, bronchovascular bundle thickening, bronchiectasis, and honeycombing), noise, artifacts, and overall image quality on a 3-point scale (1 = worst, 2 = equal, 3 = best) compared with U-HRCT
N
. Noise values were calculated quantitatively.
Results
U-HRCT could depict a 0.15-mm slit. Both U-HRCT
SHR
and U-HRCT
SHR-VOL
significantly improved visualization of normal anatomical structures and abnormal CT findings, except for intralobular reticular opacities and reduced artifacts, compared with AD-CT (
p
< 0.014). Visually, U-HRCT
SHR-VOL
has less noise than U-HRCT
SHR
and AD-CT (
p
< 0.00001). Quantitative noise values were significantly higher in the following order: U-HRCT
SHR
(mean, 30.41), U-HRCT
SHR-VOL
(26.84), AD-CT (16.03), and U-HRCT
N
(15.14) (
p
< 0.0001). U-HRCT
SHR
and U-HRCT
SHR-VOL
resulted in significantly higher overall image quality than AD-CT and were almost equal to U-HRCT
N
(
p
< 0.0001).
Conclusions
Both U-HRCT
SHR
and U-HRCT
SHR-VOL
can provide higher image quality than AD-CT, while U-HRCT
SHR-VOL
was less noisy than U-HRCT
SHR
.
Key Points
• Ultra-high-resolution CT (U-HRCT) can improve spatial resolution.
• U-HRCT can reduce streak and dark band artifacts.
• U-HRCT can provide higher image quality than conventional area detector CT.
• In U-HRCT, the volume mode is less noisy than the super-high-resolution mode.
• U-HRCT may provide more detailed information about the lung anatomy and pathology.
•Multiple abdominal organs were automatically segmented from CT images.•A method for modeling prediction-based conditional shape-location priors is developed.•Organ correlation graph is introduced, ...which embeds the spatial correlations among organs.•Proposed priors enables the estimation of intensity priors from unsupervised intensity information.•The results were comparable to state-of-the-art methods using supervised intensity information.
Display omitted
This paper addresses the automated segmentation of multiple organs in upper abdominal computed tomography (CT) data. The aim of our study is to develop methods to effectively construct the conditional priors and use their prediction power for more accurate segmentation as well as easy adaptation to various imaging conditions in CT images, as observed in clinical practice. We propose a general framework of multi-organ segmentation which effectively incorporates interrelations among multiple organs and easily adapts to various imaging conditions without the need for supervised intensity information. The features of the framework are as follows: (1) A method for modeling conditional shape and location (shape–location) priors, which we call prediction-based priors, is developed to derive accurate priors specific to each subject, which enables the estimation of intensity priors without the need for supervised intensity information. (2) Organ correlation graph is introduced, which defines how the conditional priors are constructed and segmentation processes of multiple organs are executed. In our framework, predictor organs, whose segmentation is sufficiently accurate by using conventional single-organ segmentation methods, are pre-segmented, and the remaining organs are hierarchically segmented using conditional shape–location priors. The proposed framework was evaluated through the segmentation of eight abdominal organs (liver, spleen, left and right kidneys, pancreas, gallbladder, aorta, and inferior vena cava) from 134 CT data from 86 patients obtained under six imaging conditions at two hospitals. The experimental results show the effectiveness of the proposed prediction-based priors and the applicability to various imaging conditions without the need for supervised intensity information. Average Dice coefficients for the liver, spleen, and kidneys were more than 92%, and were around 73% and 67% for the pancreas and gallbladder, respectively.
Whole-body magnetic resonance imaging (WB-MRI) is currently used worldwide for detecting bone metastases from prostate cancer. The 5-year survival rate for prostate cancer is > 95%. However, an ...increase in survival time may increase the incidence of bone metastasis. Therefore, detecting bone metastases is of great clinical interest. Bone metastases are commonly located in the spine, pelvis, shoulder, and distal femur. Bone metastases from prostate cancer are well-known representatives of osteoblastic metastases. However, other types of bone metastases, such as mixed or inter-trabecular type, have also been detected using MRI. MRI does not involve radiation exposure and has good sensitivity and specificity for detecting bone metastases. WB-MRI has undergone gradual developments since the last century, and in 2004, Takahara et al., developed diffusion-weighted Imaging (DWI) with background body signal suppression (DWIBS). Since then, WB-MRI, including DWI, has continued to play an important role in detecting bone metastases and monitoring therapeutic effects. An imaging protocol that allows complete examination within approximately 30 min has been established. This review focuses on WB-MRI standardization and the automatic calculation of tumor total diffusion volume (tDV) and mean apparent diffusion coefficient (ADC) value. In the future, artificial intelligence (AI) will enable shorter imaging times and easier automatic segmentation.
Objectives
This study was conducted in order to compare the effect of field of view (FOV) size on image quality between ultra-high-resolution CT (U-HRCT) and conventional high-resolution CT (HRCT).
...Methods
Eleven cadaveric lungs were scanned with U-HRCT and conventional HRCT and reconstructed with five FOVs (40, 80, 160, 240, and 320 mm). Three radiologists evaluated and scored the images. Three image evaluations were performed, comparing the image quality with the five FOVs with respect to the 160-mm FOV. The first evaluation was performed on conventional HRCT images, and the second evaluation on U-HRCT images. Images were scored on normal structure, abnormal findings, and overall image quality. The third evaluation was a comparison of the images obtained with conventional HRCT and U-HRCT, with scoring performed on overall image quality. Quantitative evaluation of noise was performed by setting ROIs.
Results
In conventional HRCT, image quality was improved when the FOV was reduced to 160 mm. In U-HRCT, image quality, except for noise, improved when the FOV was reduced to 80 mm. In the third evaluation, overall image quality was improved in U-HRCT over conventional HRCT at all FOVs. Noise of U-HRCT increased with respect to conventional HRCT when the FOV was reduced from 160 to 40 mm. However, at 240- and 320-mm FOVs, the noise of U-HRCT and conventional HRCT showed no differences.
Conclusions
In conventional HRCT, image quality did not improve when the FOV was reduced below 160 mm. However, in U-HRCT, image quality improved even when the FOV was reduced to 80 mm.
Key Points
• Reducing the size of the field of view to 160 mm improves diagnostic imaging quality in high-resolution CT.
• In ultra-high-resolution CT, improvements in image quality can be obtained by reducing the size of the field of view to 80 mm.
• Ultra-high-resolution CT produces images of higher quality compared with conventional HRCT irrespective of the size of the field of view.
Purpose
To compare the diagnostic performance of ring-type dedicated breast PET (dbPET), whole-body PET (WBPET), and DCE-MRI for predicting pathological complete response (pCR) after neoadjuvant ...chemotherapy (NAC).
Methods
This prospective study included 29 women with histologically proven breast cancer on needle biopsy between July 2016 and July 2019 (age: mean 55 years; range 35–78). Patients underwent WBPET followed by ring-type dbPET and DCE-MRI pre- and post-NAC for preoperative evaluation. pCR was defined as an invasive tumor that disappeared in the breast. Standardized uptake values corrected for lean body mass (SULpeak) were calculated for dbPET and WBPET scans. Maximum tumor length was measured in DCE-MRI images.
Reduction rates were calculated for quantitative evaluation. Two radiologists independently evaluated the qualitative findings. Reduction rates and qualitative findings were compared between the pCR (
n
= 7) and non-pCR (
n
= 22) groups for each modality. Differences in quantitative and qualitative data between the two groups were analyzed statistically.
Results
Significant differences were observed in the reduction rates of dbPET and DCE-MRI (
P
= 0.01 and 0.03, respectively) between the two groups. Univariate and multiple logistic regression analyses revealed that SULpeak reduction rates in WBPET and dbPET (
P
= 0.02 and
P
= 0.01, respectively) and in dbPET (odds ratio, 16.00; 95% CI 1.57–162.10;
P
= 0.01) were significant indicators associated with pCR, respectively. No between-group differences were observed in qualitative findings in the three modalities.
Conclusion
SULpeak reduction rate of dbPET > 82% was an independent indicator associated with pCR after NAC in breast cancer.
With the advent of antiretroviral therapy (ART), the prognosis of people infected with human immunodeficiency virus (HIV) has improved, and the frequency of HIV-related central nervous system (CNS) ...diseases has decreased. Nevertheless, mortality from HIV-related CNS diseases, including those associated with ART (e.g., immune reconstitution inflammatory syndrome) remains significant. Magnetic resonance imaging (MRI) can improve the outlook for people with HIV through early diagnosis and prompt treatment. For example, HIV encephalopathy shows a diffuse bilateral pattern, whereas progressive multifocal leukoencephalopathy, HIV-related primary CNS lymphoma, and CNS toxoplasmosis show focal patterns on MRI. Among the other diseases caused by opportunistic infections, CNS cryptococcosis and CNS tuberculosis have extremely poor prognoses unless diagnosed early. Immune reconstitution inflammatory syndrome shows distinct MRI findings from the offending opportunistic infections. Although distinguishing between HIV-related CNS diseases based on imaging alone is difficult, in this review, we discuss how pattern recognition approaches can contribute to their early differentiation.
PDF
