RESUMO
OBJECTIVES: To develop a pipeline for automated body composition analysis and skeletal muscle assessment with integrated quality control for large-scale application in opportunistic imaging. METHODS: First, a convolutional neural network for extraction of a single slice at the L3/L4 lumbar level was developed on CT scans of 240 patients applying the nnU-Net framework. Second, a 2D competitive dense fully convolutional U-Net for segmentation of visceral and subcutaneous adipose tissue (VAT, SAT), skeletal muscle (SM), and subsequent determination of fatty muscle fraction (FMF) was developed on single CT slices of 1143 patients. For both steps, automated quality control was integrated by a logistic regression model classifying the presence of L3/L4 and a linear regression model predicting the segmentation quality in terms of Dice score. To evaluate the performance of the entire pipeline end-to-end, body composition metrics, and FMF were compared to manual analyses including 364 patients from two centers. RESULTS: Excellent results were observed for slice extraction (z-deviation = 2.46 ± 6.20 mm) and segmentation (Dice score for SM = 0.95 ± 0.04, VAT = 0.98 ± 0.02, SAT = 0.97 ± 0.04) on the dual-center test set excluding cases with artifacts due to metallic implants. No data were excluded for end-to-end performance analyses. With a restrictive setting of the integrated segmentation quality control, 39 of 364 patients were excluded containing 8 cases with metallic implants. This setting ensured a high agreement between manual and fully automated analyses with mean relative area deviations of ΔSM = 3.3 ± 4.1%, ΔVAT = 3.0 ± 4.7%, ΔSAT = 2.7 ± 4.3%, and ΔFMF = 4.3 ± 4.4%. CONCLUSIONS: This study presents an end-to-end automated deep learning pipeline for large-scale opportunistic assessment of body composition metrics and sarcopenia biomarkers in clinical routine. KEY POINTS: ⢠Body composition metrics and skeletal muscle quality can be opportunistically determined from routine abdominal CT scans. ⢠A pipeline consisting of two convolutional neural networks allows an end-to-end automated analysis. ⢠Machine-learning-based quality control ensures high agreement between manual and automatic analysis.
Assuntos
Sarcopenia , Composição Corporal , Humanos , Músculo Esquelético/diagnóstico por imagem , Controle de Qualidade , Sarcopenia/diagnóstico por imagem , Tomografia Computadorizada por Raios X/métodosRESUMO
PURPOSE: The aim of this study was to evaluate an intravoxel incoherent motion (IVIM) model-based analysis of diffusion-weighted imaging (DWI) for assessing the response of hepatocellular carcinoma (HCC) to locoregional therapy. PATIENTS AND METHODS: Respiratory-gated DWI (b=0, 50, and 800 s/mm2) was retrospectively analyzed in 25 patients who underwent magnetic resonance imaging at 1.5 T before and 6 weeks following the first cycle of transarterial chemoembolization therapy, transarterial ethanol-lipiodol embolization therapy, and transarterial radioembolization therapy. In addition to the determination of apparent diffusion coefficient, ADC(0,800), an estimation of the diffusion coefficient, D', and the perfusion fraction, f', was performed by using a simplified IVIM approach. Parameters were analyzed voxel-wise. Tumor response was assessed in a central slice by using a region of interest (ROI) covering the whole tumor. HCCs were categorized into two groups, responders and nonresponders, according to tumor size changes on first and second follow ups (if available) and changes of contrast-enhanced region on the first follow up. RESULTS: In total, 31 HCCs were analyzed: 17 lesions were assigned to responders and 14 were to nonresponders. In responders, ADC(0,800) and D' were increased after therapy by ~30% (P=0.00004) and ~42% (P=0.00001), respectively, whereas f' was decreased by ~37% (P=0.00094). No significant changes were found in nonresponders. Responders and nonresponders were better differentiated by changes in D' than by changes in ADC(0,800) (area under the curve =0.878 vs 0.819 or 0.714, respectively). CONCLUSION: In patients with HCCs undergoing embolization therapy, diffusion changes were better reflected by D' than by conventional ADC(0,800), which is influenced by counteracting perfusion changes as assessed by f'.
RESUMO
OBJECTIVES: To compare systematically quantitative MRI, MR spectroscopy (MRS), and different histological methods for liver fat quantification in order to identify possible incongruities. METHODS: Fifty-nine consecutive patients with liver disorders were examined on a 3 T MRI system. Quantitative MRI was performed using a dual- and a six-echo variant of the modified Dixon (mDixon) sequence, calculating proton density fat fraction (PDFF) maps, in addition to single-voxel MRS. Histological fat quantification included estimation of the percentage of hepatocytes containing fat vesicles as well as semi-automatic quantification (qHisto) using tissue quantification software. RESULTS: In 33 of 59 patients, the hepatic fat fraction was >5% as determined by MRS (maximum 45%, mean 17%). Dual-echo mDixon yielded systematically lower PDFF values than six-echo mDixon (mean difference 1.0%; P < 0.001). Six-echo mDixon correlated excellently with MRS, qHisto, and the estimated percentage of hepatocytes containing fat vesicles (R = 0.984, 0.967, 0.941, respectively, all P < 0.001). Mean values obtained by the estimated percentage of hepatocytes containing fat were higher by a factor of 2.5 in comparison to qHisto. Six-echo mDixon and MRS showed the best agreement with values obtained by qHisto. CONCLUSIONS: Six-echo mDixon, MRS, and qHisto provide the most robust and congruent results and are therefore most appropriate for reliable quantification of liver fat. KEY POINTS: ⢠Six-echo mDixon correlates excellently with MRS, qHisto, and the estimated percentage of fat-containing hepatocytes. ⢠Six-echo mDixon, MRS, and qHisto provide the most robust and congruent results. ⢠Dual-echo mDixon yields systematically lower PDFF values than six-echo mDixon. ⢠The percentage of fat-containing hepatocytes is 2.5-fold higher than fat fraction determined by qHisto. ⢠Performance characteristics and systematic differences of the various methods should be considered.