Deep learning-based dominant index lesion segmentation for MR-guided radiation therapy of prostate cancer.

BACKGROUND: Dose escalation radiotherapy enables increased control of prostate cancer (PCa) but requires segmentation of dominant index lesions (DIL). This motivates the development of automated methods for fast, accurate, and consistent segmentation of PCa DIL. PURPOSE: To construct and validate a model for deep-learning-based automatic segmentation of PCa DIL defined by Gleason score (GS) ≥3+4 from MR images applied to MR-guided radiation therapy. Validate generalizability of constructed models across scanner and acquisition differences. METHODS: Five deep-learning networks were evaluated on apparent diffusion coefficient (ADC) MRI from 500 lesions in 365 patients arising from internal training Dataset 1 (156 lesions in 125 patients, 1.5Tesla GE MR with endorectal coil), testing using Dataset 1 (35 lesions in 26 patients), external ProstateX Dataset 2 (299 lesions in 204 patients, 3Tesla Siemens MR), and internal inter-rater Dataset 3 (10 lesions in 10 patients, 3Tesla Philips MR). The five networks include: multiple resolution residually connected network (MRRN) and MRRN regularized in training with deep supervision implemented into the last convolutional block (MRRN-DS), Unet, Unet++, ResUnet, and fast panoptic segmentation (FPSnet) as well as fast panoptic segmentation with smoothed labels (FPSnet-SL). Models were evaluated by volumetric DIL segmentation accuracy using Dice similarity coefficient (DSC) and the balanced F1 measure of detection accuracy, as a function of lesion aggressiveness and size (Dataset 1 and 2), and accuracy with respect to two-raters (on Dataset 3). Upon acceptance for publication segmentation models will be made available in an open-source GitHub repository. RESULTS: In general, MRRN-DS more accurately segmented tumors than other methods on the testing datasets. MRRN-DS significantly outperformed ResUnet in Dataset2 (DSC of 0.54 vs. 0.44, p < 0.001) and the Unet++ in Dataset3 (DSC of 0.45 vs. p = 0.04). FPSnet-SL was similarly accurate as MRRN-DS in Dataset2 (p = 0.30), but MRRN-DS significantly outperformed FPSnet and FPSnet-SL in both Dataset1 (0.60 vs. 0.51 [p = 0.01] and 0.54 [p = 0.049] respectively) and Dataset3 (0.45 vs. 0.06 [p = 0.002] and 0.24 [p = 0.004] respectively). Finally, MRRN-DS produced slightly higher agreement with experienced radiologist than two radiologists in Dataset 3 (DSC of 0.45 vs. 0.41). CONCLUSIONS: MRRN-DS was generalizable to different MR testing datasets acquired using different scanners. It produced slightly higher agreement with an experienced radiologist than that between two radiologists. Finally, MRRN-DS more accurately segmented aggressive lesions, which are generally candidates for radiative dose ablation.

The Effect of Stereotactic Body Radiation Therapy for Hepatocellular Cancer on Regional Hepatic Liver Function.

PURPOSE: To investigate direct radiation dose-related and inflammation-mediated regional hepatic function losses after stereotactic body radiation therapy (SBRT) in patients with hepatocellular carcinoma (HCC) and poor liver function. METHODS AND MATERIALS: Twenty-four patients with HCC enrolled on an IRB-approved adaptive SBRT trial had liver dynamic gadoxetic acid-enhanced magnetic resonance imaging and blood sample collections before and 1 month after SBRT. Gadoxetic acid uptake rate (k1) maps were quantified for regional hepatic function and coregistered to both 2-Gy equivalent dose and physical dose distributions. Regional k1 loss patterns from before to after SBRT were analyzed for effects of dose and patient using a mixed-effects model and logistic function and were associated with pretherapy liver-function albumin-bilirubin scores. Plasma levels of tumor necrosis factor α receptor 1 (TNFR1), an inflammation marker, were correlated with mean k1 losses in the lowest dose regions by Spearman rank correlation. RESULTS: The whole group had a k1 loss rate of 0.4%/Gy (2-Gy equivalent dose); however, there was a significant random effect of patient in the mixed-effect model (P < .05). Patients with poor and good liver functions lost 50% of k1 values at 12.5 and 57.2 Gy and 33% and 16% of k1 values at the lowest dose regions (<5 Gy), respectively. The k1 losses at the lowest dose regions of individual patients were significantly correlated with their TNFR1 levels after SBRT (P < .02). CONCLUSIONS: The findings suggest that regional hepatic function losses after SBRT in patients with HCC include both direct radiation dose-dependent and inflammation-mediated effects, which could influence how to manage these patients to preserve their liver function after SBRT.

Gadoxetic Acid Uptake Rate as a Measure of Global and Regional Liver Function as Compared With Indocyanine Green Retention, Albumin-Bilirubin Score, and Portal Venous Perfusion.

Purpose: Global and regional liver function assessments are important for defining the magnitude and spatial distribution of dose to preserve functional liver parenchyma and reduce incidence of hepatotoxicity from radiation therapy for intrahepatic cancer treatment. This individualized liver function-guided radiation therapy strategy is critical for patients with heterogeneous and poor liver function, often observed in cirrhotic patients treated for hepatocellular carcinoma. This study aimed to validate k1 as a measure of global and regional function through comparison with 2 well-regarded global function measures: indocyanine green retention (ICGR) and albumin-bilirubin (ALBI). Methods and Materials: Seventy-nine dynamic gadoxetic acid enhanced magnetic resonance imaging scans were acquired in 40 patients with hepatocellular carcinoma in institutional review board approved prospective protocols. Portal venous perfusion (kpv ) was quantified from gadoxetic acid enhanced magnetic resonance imaging using a dual-input 2-compartment model, and gadoxetic acid uptake rate (k1) was fitted using a linearized single-input 2-compartment model chosen for robust k1 estimation. Four image-derived measures of global liver function were tested: (1) mean k1 multiplied by liver volume (k1VL ) (functional volume), (2) mean k1 multiplied by blood distribution volume (k1Vdis ), (3) mean kpv, and (4) liver volume (VL ). The measure's correlation with corresponding ICGR and ALBI tests was assessed using linear regression. Voxel-wise similarity between k1 and kpv was compared using Spearman ranked correlation. Results: Significant correlations (P < .05) with ICGR and ALBI were found for k1VL, k1Vdis, and VL (in order of strength), but not for mean k pv . The mean ranked correlation coefficient between k1 and kpv maps was 0.09. k1 and kpv maps were predominantly mismatched in patients with poor liver function. Conclusions: The metric combining function and liver volume (k1VL ) was a stronger measure of global liver function compared with perfusion or liver volume alone, especially in patients with poor liver function. Gadoxetic acid uptake rate is promising for both global and regional liver function.

GAN and dual-input two-compartment model-based training of a neural network for robust quantification of contrast uptake rate in gadoxetic acid-enhanced MRI.

PURPOSE: Gadoxetic acid uptake rate (k1 ) obtained from dynamic, contrast-enhanced (DCE) magnetic resonance imaging (MRI) is a promising measure of regional liver function. Clinical exams are typically poorly temporally characterized, as seen in a low temporal resolution (LTR) compared to high temporal resolution (HTR) experimental acquisitions. Meanwhile, clinical demands incentivize shortening these exams. This study develops a neural network-based approach to quantitation of k1 , for increased robustness over current models such as the linearized single-input, two-compartment (LSITC) model. METHODS: Thirty Liver HTR DCE MRI exams were acquired in 22 patients with at least 16 min of postcontrast data sampled at least every 13 s. A simple neural network (NN) with four hidden layers was trained on voxel-wise LTR data to predict k1 . Low temporal resolution data were created by subsampling HTR data to contain six time points, replicating the characteristics of clinical LTR data. Both the total length and the placement of points in the training data were varied considerably to encourage robustness to variation. A generative adversarial network (GAN) was used to generate arterial and portal venous inputs for use in data augmentation based on the dual-input, two-compartment, pharmacokinetic model of gadoxetic acid in the liver. The performance of the NN was compared to direct application of LSITC on both LTR and HTR data. The error was assessed when subsampling lengths from 16 to 4 min, enabling assessment of robustness to acquisition length. RESULTS: For acquisition lengths of 16 min NRMSE (Normalized Root-Mean-Squared Error) in k1 was 0.60, 1.77, and 1.21, for LSITC applied to HTR data, LSITC applied to LTR data, and GAN-augmented NN applied to LTR data, respectively. As the acquisition length was shortened, errors greatly increased for LSITC approaches by several folds. For acquisitions shorter than 12 min the GAN-augmented NN approach outperformed the LSITC approach to a statistically significant extent, even with HTR data. CONCLUSIONS: The study indicates that data length is significant for LSITC analysis as applied to DCE data for standard temporal sampling, and that machine learning methods, such as the implemented NN, have potential for much greater resilience to shortened acquisition time than directly fitting to the LSITC model.

Quantification of liver function by linearization of a two-compartment model of gadoxetic acid uptake using dynamic contrast-enhanced magnetic resonance imaging.

Dynamic gadoxetic acid-enhanced magnetic resonance imaging (MRI) allows the investigation of liver function through the observation of the perfusion and uptake of contrast agent in the parenchyma. Voxel-by-voxel quantification of the contrast uptake rate (k1 ) from dynamic gadoxetic acid-enhanced MRI through the standard dual-input, two-compartment model could be susceptible to overfitting of variance in the data. The aim of this study was to develop a linearized, but more robust, model. To evaluate the estimated k1 values using this linearized analysis, high-temporal-resolution gadoxetic acid-enhanced MRI scans were obtained in 13 examinations, and k1 maps were created using both models. Comparison of liver k1 values estimated from the two methods produced a median correlation coefficient of 0.91 across the 12 scans that could be used. Temporally sparse clinical MRI data with gadoxetic acid uptake were also employed to create k1 maps of 27 examinations using the linearized model. Of 20 scans, the created k1 maps were compared with overall liver function as measured by indocyanine green (ICG) retention, and yielded a correlation coefficient of 0.72. In the 27 k1 maps created via the linearized model, the mean liver k1 value was 3.93 ± 1.79 mL/100 mL/min, consistent with previous studies. The results indicate that the linearized model provides a simple and robust method for the assessment of the rate of contrast uptake that can be applied to both high-temporal-resolution dynamic contrast-enhanced MRI and typical clinical multiphase MRI data, and that correlates well with the results of both two-compartment analysis and independent whole liver function measurements.