Yttrium-90 (Y-90) radioembolization is a minimally invasive treatment for liver cancer. In this method, Y-90 microspheres are injected into the hepatic artery to irradiate the tumor internally. The dosimetry methods used to prescribe the Y-90 activity to inject are neither accurate nor precise due to their simplified assumptions. For example, the nonhomogeneous distribution of Y-90 in liver observed in ex vivo samples is not accounted for in these methods. Any inaccuracy in the dosimetry can consequently affect the radioembolization effectiveness and safety. We have developed a new dosimetry based on computational fluid dynamics (CFD) simulation in the hepatic arterial tree, CFDose, to overcome some of the limitations of the current methods. However, due to the expensive computational cost of CFD simulations, this method needs to be accelerated before it can be used in real-time during treatment planning. In this paper, we introduce a convolutional neural network model trained with the CFD results of a patient with hepatocellular carcinoma to predict the Y-90 distribution under different downstream boundary conditions. For this purpose, the hepatic arterial tree of the patient was first segmented from the cone-beam computed tomography images. The CFD outlets were coupled to 3-element Windkessel circuits in order to model the downstream vasculature boundary condition. The Windkessel circuits consisted of a capacitor, proximal and distal resistor connected to a distal pressure of 15 mmHg. The arterial tree hemodynamics was then calculated by solving Navier-Stokes conservation of mass and momentum equations using an open-source software, SimVascular, for 21 different outlet conditions. The accumulated blood flow to each arterial outlet was calculated from the blood flow rate. To create a large enough dataset for the neural network training, additional flow data were interpolated. The final dataset was randomly divided into three subsets of training (60%), validation (20%), and test (20%) datasets. A neural network with 6 convolutional layers was adapted. A regression layer with a mean-squared-error loss function was used at the output of the network to estimate the Y-90 distribution between the hepatic lobe segments. Stochastic gradient decent with a momentum optimizer (momentum = 0.9) was used to solve the network. The model performance was evaluated by calculating the mean squared error and prediction accuracy. Results showed that the average difference between the actual and predicted data was less than 1%. Therefore, the proposed model could estimate the Y-90 distribution with high accuracy significantly faster than a CFD simulation. A deep neural network trained with a larger dataset may be warranted to predict the Y-90 distribution more accurately for the outlet conditions not included in the current database.


Prediction of Blood Flow Distribution in Liver Radioembolization Using Convolutional Neural Networks

Division / Sponsors: 
Aerospace, Bioengineering, Applied Mechanics, Design Engineering Division, Advanced Energy Systems, Manufacturing Engineering Division, Noise Control and Acoustics, Nondestructive Evaluation, Diagnosis, and Prognosis (NDPD), Nuclear Engineering Division, Process Industries Division, Fluids Engineering, Heat Transfer, Materials, Microelectromechanical Systems, Safety Engineering and Risk Analysis, Computers & Information in Engineering


Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2020/volume/84478

IMECE2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.