Total brain dose estimation in single-isocenter-multiple-targets (SIMT) radiosurgery via a novel deep neural network with spherical convolutions

Abstract

Background and purpose: Accurate prediction of normal brain dosimetric parameters is crucial for the quality control of single-isocenter multi-target (SIMT) stereotactic radiosurgery (SRS) treatment planning. Reliable dose estimation of normal brain tissue is one of the great indicators to evaluate plan quality and is used as a reference in clinics to improve potentially SIMT SRS treatment planning quality consistency. This study aimed to develop a spherical coordinate-defined deep learning model to predict the dose to a normal brain for SIMT SRS treatment planning.

Methods: By encapsulating the human brain within a sphere, 3D volumetric data of planning target volume (PTVs) can be projected onto this geometry as a 2D spherical representation (in azimuthal and polar angles). A novel deep learning model spherical convolutional neural network (SCNN) was developed based on spherical convolution to predict brain dosimetric evaluators from spherical representation. Utilizing 106 SIMT cases, the model was trained to predict brain V50%, V60%, and V66.7%, corresponding to V10Gy and V12Gy, as key dosimetric indicators. The model prediction performance was evaluated using the coefficient of determination (R²), mean absolute error (MAE), and mean absolute percentage error (MAPE).

Results: The SCNN accurately predicted normal brain dosimetric values from the modeled spherical PTV representation, with R² scores of 0.92 ± 0.05/0.94 ± 0.10/0.93 ± 0.09 for V50%/V60%/V66.7%, respectively. MAEs values were 1.94 ± 1.61 cc/1.23 ± 0.98 cc/1.13 ± 0.99 cc, and MAPEs were 19.79 ± 20.36%/20.79 ± 21.07%/21.15 ± 22.24%, respectively.

Conclusions: The deep learning model provides treatment planners with accurate prediction of dose to normal brain, enabling improved consistency in treatment planning quality. This method can be extended to other brain-related analyses as an efficient data dimension reduction method.

Keywords: SIMT; SRS; deep learning; plan quality; spherical convolution.

Figure 1.

(A) The conceptual design of our proposed spherical projection; (B) The demonstration of the ray tracing-based spherical projection. The blue arrows indicate the high-intensity regions after the projection. (C) The distribution of the fitted radii.

Figure 2,

(A) The Equal Area isoLatitude Pixelisation (HEALPix); (B) The overall design of the proposed spherical convolutional neural network (SCNN) model for V50%, V60%, and V66.7% prediction.

Figure 3.

The proposed spherical data augmentation technique.

Figure 4.

The design of three comparison models. (A) classic 2D U-Net encoder with unwrapped spherical projections as input; (B) classic 2D U-Net encoder with planar projections as input; (C) classic 3D U-Net encoder with original 3D PTV as input.

Figure 5.

The absolute and percentage errors (obtained from our SCNN model) relative to the ground truth V50%, V60%, and V66.7% values.

Figure 6.

An example of using the V60% prediction for suboptimal plan identification. (A)-(B): original SIMT plan in two axial views; (C)-(D) improved SIMT plan in the corresponding axial views after V50%/V60%/V66.7% prediction-guided replanning. The blue arrows show improved in-plane V12Gy (white lines) conformity, and the red arrows show improved superior-inferior V4Gy (red lines) conformity.