Patient scans from MRI often suffer from noise, which hampers the diagnostic capability of such images. As a method to mitigate such artifact, denoising is largely studied both within the medical imaging community and beyond the community as a general subject. However, recent deep neural network-based approaches mostly rely on the minimum mean squared error (MMSE) estimates, which tend to produce a blurred output. Moreover, such models suffer when deployed in real-world sitautions: out-of-distribution data, and complex noise distributions that deviate from the usual parametric noise models. In this work, we propose a new denoising method based on score-based reverse diffusion sampling, which overcomes all the aforementioned drawbacks. Our network, trained only with coronal knee scans, excels even on out-of-distribution in vivo liver MRI data, contaminated with complex mixture of noise. Even more, we propose a method to enhance the resolution of the denoised image with the same network. With extensive experiments, we show that our method establishes state-of-the-art performance, while having desirable properties which prior MMSE denoisers did not have: flexibly choosing the extent of denoising, and quantifying uncertainty.
Considerable interobserver variability exists in the assignment at CT of local resectability of pancreatic cancer, even amongst experienced radiologists. Background Accurate assessment of local resectability of pancreatic cancer at initial workup is critical to determine the most appropriate management strategy among up-front operation, neoadjuvant ...
A risk score system based on attenuation coefficient and two-dimensional shear-wave elastography showed the best diagnostic performance for nonalcoholic steatohepatitis compared with serum markers ... Background Multiparametric US examination may have potential in the comprehensive evaluation of nonalcoholic fatty liver disease (NAFLD), but multicenter studies are lacking. Purpose To evaluate th...
This study was aimed to investigate the value of quantitative ultrasound (US) parameters from radiofrequency (RF) data analysis for assessing hepatic steatosis, using controlled attenuation parameter (CAP)-based steatosis grades as the reference standard. We analyzed 243 participants with both B-mode liver US with RF data acquisition and CAP measurements. On B-mode US images, hepatic steatosis was visually scored (0/1/2/3, none/mild/moderate/severe), and the hepatorenal index (HRI) was calculated. From the RF data analysis, the tissue scatter-distribution imaging parameter (TSI-p) and tissue attenuation imaging parameter (TAI-p) of the liver parenchyma were measured. US parameters were correlated with CAP-based steatosis grades (S0/1/2/3, none/mild/moderate/severe) and their diagnostic performance was evaluated using receiver operating characteristic (ROC) curve analysis. Multivariate linear regression analysis was performed to identify determinants of TSI-p and TAI-p. Participants were classified as having S0 (n=152), S1 (n=54), S2 (n=14), and S3 (n=23) on CAP measurements. TSI-p and TAI-p were significantly correlated with steatosis grades (ρ =0.593 and ρ=-0.617, P<0.001 for both). For predicting ≥S1, ≥S2, and S3, the areas under the ROC curves (AUCs) of TSI-p were 0.827/0.914/0.917; TAI-p, 0.844/0.914/0.909; visual scores, 0.659/0.778/0.794; and HRI, 0.629/0.751/0.759, respectively. TSI-p and TAI-p had significantly higher AUCs than did visual scores or HRI for ≥S1 or ≥S2 (P≤0.003). In the multivariate analysis, the transient elastography-based fibrosis grade (P=0.034) and steatosis grade (P<0.001) were independent determinants of TSI-p, while steatosis grade (P<0.001) was an independent determinant of TAI-p. TSI-p and TAI-p derived from US RF data may be useful for detecting hepatic steatosis and assessing its severity.