Logo: to the web site of Uppsala University

Background and purpose:

Devices that combine an MR-scanner with a Linac for radiotherapy, referred to as MR-Linac systems, introduce the possibility to acquire high resolution images prior and during treatment. Hence, there is a possibility to acquire individualised learning sets for motion models for each fraction and the construction of intrafractional motion models. We investigated the feasibility for a principal component analysis (PCA) based, intrafractional motion model of the male pelvic region.

Materials and methods:

4D-scans of nine healthy male volunteers were utilized, FOV covering the entire pelvic region including prostate, bladder and rectum with manual segmentation of each organ at each time frame. Deformable image registration with an optical flow algorithm was performed for each subject with the first time frame as reference. PCA was performed on a subset of the resulting displacement vector fields to construct individualised motion models evaluated on the remaining fields.

Results:

The registration algorithm produced accurate registration result, in general DICE overlap >0.95 across all time frames. Cumulative variance of the eigen values from the PCA showed that 50% or more of the motion is explained in the first component for all subjects. However, the size and direction for the components differed between subjects. Adding more than two components did not improve the accuracy significantly and the model was able to explain motion down to about 1 mm.ConclusionsAn individualised intrafractional male pelvic motion model is feasible. Geometric accuracy was about 1 mm based on 1–2 principal components.

Background and Purpose: Treatments on combined Magnetic Resonance (MR) scanners and Linear Accelerators (Linacs) for radiotherapy, called MR-Linacs, often require daily contouring. Currently, deformable image registration (DIR) algorithms propagate contours from reference scans, however large shape and size changes can be troublesome. Artificial neural network (ANN) based contouring may alleviate this issue, however generally requires large datasets for training. Mitigating the problem of scarcity of data, we propose patient specific networks trained on a single dataset for each patient, for contouring onto the following datasets in an adaptive MR-Linacworkflow. Materials and Methods: MR-scans from 17 prostate patients treated on an MR-Linac with contours of Clinical Target Volume (CTV), bladder and rectum were utilized. U-net shaped models were trained based on the image from the first fraction of each patient, and subsequently applied onto the following treatment images. Results were compared with manual contours in terms of the Dice coefficient and Added Path Length (APL). As benchmark, contours propagated through the clinical DIR algorithm were similarly evaluated. Results: In Dice coefficient the ANN output was 0.92 +/- 0.03, 0.93 +/- 0.07 and 0.84 +/- 0.10 while for DIR 0.95 +/- 0.03, 0.93 +/- 0.08, 0.88 +/- 0.06 for CTV, bladder and rectum respectively. Similarly, APL where 3109 +/- 1642, 7250 +/- 4234 and 5041 +/- 2666 for ANN and 1835 +/- 1621, 7236 +/- 4287 and 4170 +/- 2920 voxels for DIR. Conclusions: Patient specific ANN models trained on images from the first fraction of a prostate MR-Linac treatment showed similar accuracy when applied to the subsequent fraction images as a clinically implemented DIR method.

  Background

Daily adaptive radiotherapy, as performed with the Elekta Unity MR-Linac, requires choosing between different adaptation methods, namely ATP (Adapt to Position) and ATS (Adapt to Shape), where the latter requires daily re-contouring to obtain a dose plan tailored to the daily anatomy. These steps are inherently resource-intensive, and quickly predicting the dose distribution and the dosimetric evaluation criteria while the patient is on the table could facilitate a fast selection of adaptation method and decrease the treatment times.

Purpose

In this work, we aimed to develop a deep-learning-based dose-prediction pipeline for prostate MR-Linac treatments.

Methods

Two hundred twelve MR-images, structure sets, and dose distributions from 35 prostate patients treated with 6.1 Gy for 7 or 6 fractions at our MR-Linac were included, split into train/test partitions of 152/60 images, respectively. A deep-learning segmentation network was trained to segment the CTV (prostate), bladder, and rectum. A second network was trained to predict the dose distribution based on manually delineated structures. At inference, the predicted segmentations acted as input to the dose prediction network, and the predicted dose was compared to the true (optimized in the treatment planning system) dose distribution.

Results

Median DSC values from the segmentation network were 0.90/0.94/0.87 for CTV/bladder/rectum. Predicted segmentations as input to the dose prediction resulted in mean differences between predicted and true doses of 0.7%/0.7%/1.7% (relative to the prescription dose) for D_98%/D_95%/D_2% for the CTV. For the bladder, the difference was 0.7%/0.3% for D_mean/D_2% and for the rectum 0.1/0.2/0.2 pp (percentage points) for V_33Gy/V_38Gy/V_41Gy. In comparison, true segmentations as input resulted in differences of 1.1%/0.9%/1.6% for CTV, 0.5%/0.4% for bladder, and 0.7/0.4/0.3 pp for the rectum. Only D_2% for CTV and D_mean/D_2% for bladder were found to be statistically significantly better when using true structures instead of predicted structures as input to the dose prediction.

Conclusions

Small differences in the fulfillment of clinical dose-volume constraints are seen between utilizing deep-learning predicted structures as input to a dose prediction network and manual structures. Overall mean differences <2% indicate that the dose-prediction pipeline is useful as a decision support tool where differences are >2%.