Deep Learning Combined with Radiologist's Intervention Achieves Accurate Segmentation of Hepatocellular Carcinoma in Dual-Phase Magnetic Resonance Images.

Purpose: Segmentation of hepatocellular carcinoma (HCC) is crucial; however, manual segmentation is subjective and time-consuming. Accurate and automatic lesion contouring for HCC is desirable in clinical practice. In response to this need, our study introduced a segmentation approach for HCC combining deep convolutional neural networks (DCNNs) and radiologist intervention in magnetic resonance imaging (MRI). We sought to design a segmentation method with a deep learning method that automatically segments using manual location information for moderately experienced radiologists. In addition, we verified the viability of this method to assist radiologists in accurate and fast lesion segmentation. Method: In our study, we developed a semiautomatic approach for segmenting HCC using DCNN in conjunction with radiologist intervention in dual-phase gadolinium-ethoxybenzyl-diethylenetriamine penta-acetic acid- (Gd-EOB-DTPA-) enhanced MRI. We developed a DCNN and deep fusion network (DFN) trained on full-size images, namely, DCNN-F and DFN-F. Furthermore, DFN was applied to the image blocks containing tumor lesions that were roughly contoured by a radiologist with 10 years of experience in abdominal MRI, and this method was named DFN-R. Another radiologist with five years of experience (moderate experience) performed tumor lesion contouring for comparison with our proposed methods. The ground truth image was contoured by an experienced radiologist and reviewed by an independent experienced radiologist. Results: The mean DSC of DCNN-F, DFN-F, and DFN-R was 0.69 ± 0.20 (median, 0.72), 0.74 ± 0.21 (median, 0.77), and 0.83 ± 0.13 (median, 0.88), respectively. The mean DSC of the segmentation by the radiologist with moderate experience was 0.79 ± 0.11 (median, 0.83), which was lower than the performance of DFN-R. Conclusions: Deep learning using dual-phase MRI shows great potential for HCC lesion segmentation. The radiologist-aided semiautomated method (DFN-R) achieved improved performance compared to manual contouring by the radiologist with moderate experience, although the difference was not statistically significant.

Comparing and combining MRE, T1ρ, SWI, IVIM, and DCE-MRI for the staging of liver fibrosis in rabbits: Assessment of a predictive model based on multiparametric MRI.

PURPOSE: To establish and validate an optimal predictive model based on multiparametric MRI for staging liver fibrosis (LF) in rabbits with magnetic resonance elastography (MRE), spin-lattice relaxation time in the rotating frame (T1ρ imaging), SWI, intravoxel incoherent motion (IVIM), and DCE-MRI. METHODS: The LF group included 120 rabbits induced by subcutaneous injections of carbon tetrachloride (CCl4 ); 30 normal rabbits served as the control group. Multiparametric MRI was performed, including MRE, T1ρ, SWI, IVIM, and DCE-MRI. The quantitative parameters were analyzed in two groups, with histopathological results serving as the reference standard. The diagnostic performance of multiparametric MRI and the predictive model established by multivariable logistic regression analysis were evaluated by receiver operating characteristic (ROC) curve analysis. RESULTS: In total, 32, 67, and 51 rabbits were histologically diagnosed as no fibrosis (stage F0), early-stage LF (F1-F2), and advanced-stage LF (F3-F4), respectively. The LF stages presented a strong correlation with liver stiffness (LS) on MRE (r = 0.90), signal-intensity ratio (SIR) on SWI (r = -0.84), and Ktrans on DCE-MRI (r = 0.71; p < 0.05 for all). The LS and SIR parameters had higher AUC values for distinguishing early-stage LF from both no fibrosis (0.94 and 0.93, respectively) and advanced-stage LF (0.95 and 0.87, respectively). The predictive model showed a slightly higher AUC value of 0.97 (0.90-0.99) than LS and SIR in distinguishing early-stage LF from no fibrosis (p > 0.05), a significantly higher AUC value of 0.98 (0.93-0.99) than the SIR in distinguishing early-stage from advanced-stage LF (p < 0.05). CONCLUSION: SWI, DCE-MRI, and MRE in particular showed improved performance for LF diagnosis and stage. The predictive model based on multiparametric MRI was found to further enhance diagnostic accuracy and could serve as an excellent imaging tool for staging LF.

Dynamic Contrast-enhanced Magnetic Resonance Imaging for Differentiating Between Primary Tumor, Metastatic Node and Normal Tissue in Head and Neck Cancer.

OBJECTIVE: To study the difference of the Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) parameters among the primary tumor, metastatic node and peripheral normal tissue of head and neck cancer. MATERIALS AND METHODS: Consecutive newly-diagnosed head and neck cancer patients with nodal metastasis between December 2010 and July 2013 were recruited, and 25 patients (8 females; 24~63,mean 43±11 years old) were enrolled. DCE-MRI was performed in the primary tumor region including the regional lymph nodes on a 3.0-T MRI system. Three quantitative parameters: Ktrans (volume transfer constant), ve (volume fraction of extravascular extracellular space) and kep (the rate constant of contrast transfer) were calculated for the largest node. A repeated-measure ANOVA with a Greenhouse-Geisser correction and post hoc tests using the Bonferroni correction were used to evaluate the differences in Ktrans, ve and kep among primary tumors, metastatic nodes and normal tissue. RESULTS: The values of both Ktrans and ve of normal tissue differed significantly from those of nodes (both P < 0.001) and primary tumors (both P < 0.001) respectively, while no significant differences of Ktrans and ve were observed between nodes and primary tumors (P = 0.075 and 0.365 respectively). The kep values of primary tumors were significantly different from those of nodes (P = 0.001) and normal tissue (P = 0.002), while no significant differences between nodes and normal tissue (P > 0.999). CONCLUSION: The DCE-MRI parameters were different in the tumors, metastatic nodes and normal tissue in head and neck cancer. These findings may be useful in the characterization of head and neck cancer.

Discriminating Value of Calprotectin in Disease Activity and Progression of Nonradiographic Axial Spondyloarthritis and Ankylosing Spondylitis.

It has been controversial whether ankylosing spondylitis (AS) and nonradiographic axial spondyloarthritis (nr-axSpA) are separate or different phases of radiographic progression. We determined that serum calprotectin level (ng/ml) was higher in AS (15.30 ± 6.49) and nr-axSpA (17.76 ± 8.59) patients than in healthy individuals (7.40 ± 2.67). No difference was observed in calprotectin level between these two groups. Elevated calprotectin was positively correlated with ESR, CRP, BASDAI, and ASDAS as well as SPARCC scoring and had no correlation with BASFI and mSASSS. No correlation was observed between calprotectin and Wnt/ß-catenin pathway markers. Serum calprotectin can be used as a marker for inflammation in both nr-axSpA and AS, while it does not contribute to the discrimination of AS and nr-axSpA. Calprotectin-mediated inflammation was not correlated with principle effectors of Wnt/ß-catenin pathway, indicating that inflammation and bone fusion might be separate processes of the disease.

Magnetic resonance elastography in a rabbit model of liver fibrosis: a 3-T longitudinal validation for clinical translation.

This study aimed to determine the relationships between magnetic resonance elastography (MRE) imaging biomarkers and the stages of liver fibrosis in a rabbit model of liver fibrosis, a longitudinal validation for clinical translation. Liver fibrosis was induced in 38 male New Zealand rabbits by weekly subcutaneous injections of 0.1 ml 50% carbon tetrachloride oily solution per kilogram of body weight for 4 to 10 weeks to produced varying degrees of liver fibrosis. The values for the liver stiffness (LS) MRE imaging biomarkers were measured at different stages of liver fibrosis. Masson trichrome staining of liver tissue was used to identify collagen tissue. Among the 38 rabbits, the histological studies showed liver fibrosis stage 1 (F1, n = 11), liver fibrosis stage 2 (F2, n = 8), liver fibrosis stage 3 (F3, n = 7), and liver fibrosis stage 4 (F4, liver cirrhosis, n = 12). Additional healthy rabbits served as controls (F0, n = 15). During liver fibrosis progression, the mean LS values increased during liver fibrosis progression. There were significant differences in LS values between (F0 and F1) and (F2 and F3), (F2 and F3) and (F4), and (F0 and F1) and (F4), which are three clinically relevant fibrosis groups. There was a high correlation between the LS values measured by MRE and the stages of liver fibrosis determined by histology (R2 = 0.67, P < 0.001). MRE imaging has the potential to serve as a noninvasive, unenhanced imaging technique for liver fibrosis diagnosis and staging.

[Value of susceptibility weighted imaging in hepatic fibrosis staging by using MR in a rabbit model].

OBJECTIVE: To assess the feasibility of susceptibility weighted imaging (SWI) in staging hepatic fibrosis(HF). METHODS: Sixty healthy rabbits were divided into three groups: HF group(n=32), control group(n=16), supplementary group(n=12). Rabbits in HF group and supplementary group were injected subcutaneously with 50% CCl4 oil solution to establish hepatic fibrosis model. On the basis of preliminary test, eight rabbits from HF group and four rabbits from control group underwent liver conventional MR scans and SWI once a time at 4, 5, 6, 10 weeks after CCl4 administration.After MR scans at each time point, rabbits were killed to detect pathological staging with Scheuer staging.The liver signal intensity (SI) and liver-to-muscle SI ratios (SIR) were measured. According to the Scheuer classification of histological fibrosis stages, the correlation about the SI value, SIR value and the histological fibrosis stages was investigated by using the Spearman correlation test. The receiver operating characteristic (ROC) curve was used to evaluate the diagnostic performance of SWI for staging HF on the basis of the histopathologic analysis of HF. RESULTS: There were fifteen rabbits in pathological staging F0, the value of SIR and SI was 0.974 ± 0.018, 374±44, SIR values of pathological staging from F1 to F4 were 0.963 ± 0.018, 0.796 ± 0.023, 0.786 ± 0.025, 0.512±0.024 respectively. SI values of pathological staging from F1 to F4 were 372±18, 376±22, 346±15, 288±19 respectively. In the early period of liver fibrosis, there were no statistical differences in the SI value between F0 and F1, F1 and F2 stage.With progression of hepatic fibrosis, from F2 to F4, SI value decreased, the difference was statistically significant (P<0.05). With the progress of liver fibrosis, SIR value was reduced. It was negatively correlated with the HF stages and SIR value(r=-0.896, P<0.05). ROC curve analysis showed that the efficiency of SI value diagnosis in liver fibrosis was high in the late stage of liver fibrosis, but it was low in the early stage.The performance of liver-to-muscle SI ratio on SWI was high in the early stage. Liver-to-muscle SI ratio had a higher diagnostic performance than SI in the diagnosis of liver fibrosis stages. CONCLUSION: SWI can be a safe, reliable method for staging hepatic fibrosis and provide quantitative imaging basis for clinical treatment.

An experimental study on the assessment of rabbit hepatic fibrosis by using magnetic resonance T1ρ imaging.

OBJECTIVE: To explore the correlation between the T1ρ values of liver and liver fibrosis by using magnetic resonance (MR) T1ρ imaging. MATERIALS AND METHODS: The study consisted of the control group and the hepatic fibrosis (HF) group. Carbon tetrachloride (CCl4) injection was performed once a week for 10 weeks (week 1-10) in the HF group which was divided into 5 subgroups and underwent MR examinations at weeks 4, 5, 6, 10, and 15 respectively post the first CCl4 injection (week 1). According to Scheuer Classification, the stage of HF of all rabbits was classified as S0-S4. Mann-Whitney U test was performed to compare the T1ρ values, and p<0.05 was considered statistically significant. RESULTS: The control group included 11 rabbits and the HF group included 46 rabbits. The T1ρ values in the HF group tended to increase with the increase of CCl4 injection duration or in higher HF stages. The T1ρ values were significantly lower (p=0.036) in the control group (or the stage S0 group) (23.5±4.0 ms) compared to the whole HF group (or S1-S4 group) (26.1±3.7 ms), and dropped at week 15 (p=0.043) after the CCl4 injection was stopped at week 10. CONCLUSION: T1ρ imaging is closely associated with the severity of HF and may play an important role in the early diagnosis of HF.

Comparison of magnetic resonance elastography and diffusion-weighted imaging for staging hepatic fibrosis.

BACKGROUND: To compare the diagnostic values of magnetic resonance elastography (MRE) and diffusion-weighted imaging (DWI) in staging hepatic fibrosis (HF) in an animal model. METHODS: This study consisted of 44 rabbits served as HF group and 9 normal rabbits. HF group was divided into two subgroups: Group A (n = 32) and Group B (n = 12). Rabbits in Group B were served as a complementary group when rabbits in Group A suddenly died during the study. Rabbits from control and Group A underwent abdominal MR imaging (MRI), MRE, and DWI. In Group A, random eight rabbits underwent MRI examinations at 4, 5, 6, 10 weeks after carbon tetrachloride oil subcutaneous injection. Liver stiffness (LS) and apparent diffusion coefficient (ADC) values of liver parenchyma were measured. The diagnostic performance of MRE and DWI for staging HF was compared using the receiver operating characteristic curve analysis on the basis of the histopathological analysis of HF. RESULTS: Significant differences of LS and DWI values were present among HF stages (P < 0.005). The LS values measured on MRE (r = 0.838, P < 0.001) were more strongly correlated with the HF stages than with ADC values (r = -0.527, P < 0.001). The area under the receiver operating characteristic curve values of LS were significantly larger than those of DWI were for discriminating two stages of HF (0.979 vs. 0.712 for ≥ S1, 0.922 vs. 0.699 for ≥ S2). MRE showed higher specificity for predicting all stages of HF compared to DWI. CONCLUSIONS: MRE more strongly correlated with the HF stages than DWI and is more specific in predicting all HF stages.

Tumor volumes measured from static and dynamic 18F-fluoro-2-deoxy-D-glucose positron emission tomography-computed tomography scan: comparison of different methods using magnetic resonance imaging as the criterion standard.

OBJECTIVE: The objective of this study was to compare the accuracy of calculating the primary tumor volumes using a gradient-based method and fixed threshold methods on the standardized uptake value (SUV) maps and the net influx of FDG (Ki) maps from positron emission tomography-computed tomography (PET-CT) images. MATERIALS AND METHODS: Newly diagnosed patients with head and neck cancer were recruited, and dynamic PET-CT scan and T2-weighted magnetic resonance imaging were performed. The maps of Ki and SUV were calculated from PET-CT images. The tumor volumes were calculated using a gradient-based method and a fixed threshold method at 40% of maximal SUV or maximal Ki. Four kinds of volumes, VOLKi-Gra (from the Ki maps using the gradient-based method), VOLKi-40% (from the Ki maps using the threshold of 40% maximal Ki), VOLSUV-Gra (from the SUV maps using the gradient-based method), and VOLSUV-40% (from the SUV maps using the threshold of 40% maximal SUV), were acquired and compared with VOLMRI (the volumes acquired on T2-weighted images) using the Pearson correlation, paired t test, and similarity analysis. RESULTS: Eighteen patients were studied, of which 4 had poorly defined tumors (PDT). The positron emission tomography-derived volumes were as follows: VOLSUV-40%, 2.1 to 41.2 cm (mean [SD], 12.3 [10.6]); VOLSUV-Gra, 2.2 to 28.1 cm (mean [SD], 13.2 [8.4]); VOLKi-Gra, 2.4 to 17.0 cm (mean [SD], 9.5 [4.6]); and VOLKi-40%, 2.7 to 20.3 cm (mean [SD], 12.0 [6.0]). The VOLMRI ranged from 2.9 to 18.1 cm (mean [SD], 9.1 [3.9]). The VOLKi-Gra significantly correlated with VOLMRI with the highest correlation coefficient (PDT included, R = 0.673, P = 0.002; PDT excluded, R = 0.841, P < 0.001) and presented no difference from VOLMRI (P = 0.672 or 0.561, respectively, PDT included and excluded). The difference between VOLKi-Gra and VOLMRI was also the smallest. CONCLUSIONS: The tumor volumes delineated on the Ki maps using the gradient-based method are more accurate than those on the SUV maps and using the fixed threshold methods.