Knowledge‐based isocenter selection in radiosurgery planning

Abstract

We present a new method for knowledge-based isocenter selection for treatment planning in radiosurgery. Our objective is to develop a prediction model that can learn from past manually designed treatment plans. We leverage recent advances in deep learning to predict isocenter locations in treatment plans in order to provide a decision support tool. Methods: The proposed method adapts a geometric approach using orthogonal moment expansions as a feature vector for describing the shape of the tumor. Our approach accounts primarily for tumor shape and OAR proximity, the two factors that are known to greatly affect the isocenter placement. We solve the prediction problem by training a residual neural network with skip connections on the formed shape descriptors. Our network was trained on 533 patient cases and was validated on a set of out-of-sample cases. Results: Our method generates heatmap predictions for isocenter locations that are in most cases comparable to the experienced human planners, which shows that the method can be used in treatment planning to guide the users for determining the isocenters. Conclusions: Our numerical experiments indicate a positive predictive value on an independent validation set when compared against a test dataset that was not seen by the model during training.

Knowledge‐based planning for Gamma Knife

“…As one of the main goals of KBP is to streamline planning, additional testing of the pipelines should be done using algorithmically selected isocenters. 27 Third, the efficacy of the pipeline was tested on only 27 patients due to the difficulty in obtaining patient data, particularly the dose kernels. Since the GK population is heterogeneous, with different diagnoses, target sizes, prescription doses, and target shapes, testing the pipeline on a larger sample of…”

Impact of ovary-sparing treatment planning on plan quality, treatment time and gamma passing rates in intensity-modulated radiotherapy for stage I/II cervical cancer

“…To improve the motion range of the MLC, all treatment centers and weight point locations were set at the geometric center of the PTV. [26][27][28] According to the technical conditions of our department, the prescription dose was set relatively high, [29] external radiation was 45 Gy/25 f assisted 5-7 times of 192 Ir afterloading treatment, the fractional dose of afterloading treatment was 5-6 Gy, and the equivalent dose in 2 Gy (EQD2) of external irradiation plus internal irradiation was more than 80 Gy.…”

Improved $α$-GAN architecture for generating 3D connected volumes with an application to radiosurgery treatment planning

“…One specific use case for synthetic 3D image generation is radiosurgery treatment planning, where the objective is to find a set of isocenters (i.e., focus points for radiation beams) and associated radiation amounts so that tumor volumes are eradicated with minimum dosage to surrounding healthy tissues. Berdyshev et al (2020) proposed ResNet models to learn from existing brain tumor data and previous treatment plans, and the authors noted the need for larger datasets for a high-performance automated ML-based treatment planning. Additionally, data privacy concerns prevent making such datasets publicly available, and synthetic data generation can offer a remedy for this issue.…”

“…The GANs' capabilities in synthesizing 3D images have increased their potential in medical applications, such as generating more diverse and unseen 3D brain MRI data to address clinically difficult tasks (Hong et al, 2021;Chong and Ho, 2021;Kwon et al, 2019). Another significant application of GANs in the medical domain is generating treatment planning schemes to help with knowledge-based planning and treatment planning automation (Berdyshev et al, 2020;Cevik et al, 2018). Babier et al (2020) proposed a 3D GAN image-to-image translation structure adapted from Pix2pix (Isola et al, 2017) to solve a dose distribution problem.…”

“…Additionally, their method outperforms a similar 2D GAN structure applied on 2D slices of the image (Mahmood et al, 2018), confirming that the correlations between 2D slices offer useful information. Berdyshev et al (2020) specifically focused on predicting the locations of isocenters inside the tumor volumes, which is an important step in radiosurgery treatment planning. They proposed a dimensionality reduction technique for the 3D tumor shapes to avoid the curse of dimensionality and employed a residual neural network for the isocenter prediction task.…”