Deformable registration of magnetic resonance images using unsupervised deep learning in neuro-/radiation oncology.

PURPOSE: Accurate deformable registration of magnetic resonance imaging (MRI) scans containing pathologies is challenging due to changes in tissue appearance. In this paper, we developed a novel automated three-dimensional (3D) convolutional U-Net based deformable image registration (ConvUNet-DIR) method using unsupervised learning to establish correspondence between baseline pre-operative and follow-up MRI scans of patients with brain glioma. METHODS: This study involved multi-parametric brain MRI scans (T1, T1-contrast enhanced, T2, FLAIR) acquired at pre-operative and follow-up time for 160 patients diagnosed with glioma, representing the BraTS-Reg 2022 challenge dataset. ConvUNet-DIR, a deep learning-based deformable registration workflow using 3D U-Net style architecture as a core, was developed to establish correspondence between the MRI scans. The workflow consists of three components: (1) the U-Net learns features from pairs of MRI scans and estimates a mapping between them, (2) the grid generator computes the sampling grid based on the derived transformation parameters, and (3) the spatial transformation layer generates a warped image by applying the sampling operation using interpolation. A similarity measure was used as a loss function for the network with a regularization parameter limiting the deformation. The model was trained via unsupervised learning using pairs of MRI scans on a training data set (n = 102) and validated on a validation data set (n = 26) to assess its generalizability. Its performance was evaluated on a test set (n = 32) by computing the Dice score and structural similarity index (SSIM) quantitative metrics. The model's performance also was compared with the baseline state-of-the-art VoxelMorph (VM1 and VM2) learning-based algorithms. RESULTS: The ConvUNet-DIR model showed promising competency in performing accurate 3D deformable registration. It achieved a mean Dice score of 0.975 ± 0.003 and SSIM of 0.908 ± 0.011 on the test set (n = 32). Experimental results also demonstrated that ConvUNet-DIR outperformed the VoxelMorph algorithms concerning Dice (VM1: 0.969 ± 0.006 and VM2: 0.957 ± 0.008) and SSIM (VM1: 0.893 ± 0.012 and VM2: 0.857 ± 0.017) metrics. The time required to perform a registration for a pair of MRI scans is about 1 s on the CPU. CONCLUSIONS: The developed deep learning-based model can perform an end-to-end deformable registration of a pair of 3D MRI scans for glioma patients without human intervention. The model could provide accurate, efficient, and robust deformable registration without needing pre-alignment and labeling. It outperformed the state-of-the-art VoxelMorph learning-based deformable registration algorithms and other supervised/unsupervised deep learning-based methods reported in the literature.

On the Feasibility of Stereotactic Radiosurgery With 5.0 and 10.0 mm MLC Leaves as a Function of Target Size and Shape.

Introduction: This study explores the feasibility of SRS/SRT treatment with MLC leaves wider than 2.5 mm at isocenter by inter-comparing treatment plans produced with 2.5, 5.0, and 10.0 mm leaves for various target sizes and shapes. Materials and methods: Forty previously treated patients were re-planned using 2.5, 5.0, and 10.0 mm wide MLC leaves. For each patient, all three plans were evaluated and contrasted between them in terms of five metrics: target dose homogeneity, conformity index, organs at risk dose, dose fall off outside the target, and dose to normal tissues. A regularity index RI was introduced that quantified the degree of target shape irregularity. The effect of target size and shape irregularity on feasibility of 5.0 and 10.0 mm leaves was analyzed. Results: Consistent plan degradation was observed for 10.0 mm (sometimes for 5.0 mm) compared to 2.5 mm MLC in terms of the above five plan metrics, but this degradation was small to clinically insignificant. As an exception, when target (PTV) size was smaller than about 1 cm diameter, clinically significant differences were found between 2.5, 5.0, and 10.0 mm MLC. Conclusion: 5.0 and 10.0 mm MLC can be used in SRS/SRT for targets (PTV) diameter larger than 1 cm. For smaller targets, 2.5 mm MLC is clinically superior, 5.0 is acceptable and 10.0 mm MLC is discouraged in terms of PTV dose conformity.

Predictors of Cardiac Sparing in Deep Inspiration Breath-Hold for Patients With Left Sided Breast Cancer.

Purpose: The purpose of this study was to evaluate patient-related non-dosimetric predictors of cardiac sparing with the use of deep inspiration breath-hold (DIBH) in patients with left-sided breast cancer undergoing irradiation (RT). Materials and Methods: We retrospectively reviewed charts and treatment plans of one-hundred and three patients with left-sided breast cancer. All patients had both free-breathing (FB) and DIBH (with body surface tracking) plans available. (MHD) and V4 (heart volume receiving at least 4 Gy) were extracted from dose volume histograms. Fisher's exact and Chi-square tests were used to identify predictors of reductions in MHD and V4 after DIBH. Results: One-hundred and three patients were identified and most underwent mastectomy. MHD and V4 decreased significantly in DIBH plans (0.74 ± 0.25 Gy vs. 1.72 ± 0.98 Gy, p < 0.0001 for MHD; 4 ± 4.98 cc vs. 20.79 ± 18.2 cc, p < 0.0001 for V4). Body mass index (BMI), smoking and timing of CT simulation (spring/winter vs. summer/fall) were significant predictors of reduction in MHD whereas BMI, field size, chemotherapy, axillary dissection, and timing of CT simulation predicted reduction in V4. On multivariate analysis, BMI, and timing of CT simulation remained significant predictors of the heart-sparing effect of DIBH. Conclusions: In the setting of limited resources, identifying patients who will benefit the most from DIBH is extremely important. Prior studies have identified multiple dosimetric predictors of cardiac sparing and hereby we identified new non-dosimetric factors such as BMI and timing of treatments.

A simple technique for an accurate shielding of the lungs during total body irradiation.

PURPOSE: During total body irradiation (TBI), customized shielding blocks are positioned in front of the lungs to reduce radiation dose. The difficulty is to accurately position the blocks to cover the entire lungs. A new technique based on Computed Tomography (CT) simulation was developed to determine the exact position of lung blocks prior to treatment in order to decrease overall treatment time and reduce patient discomfort. MATERIAL/METHODS: Patients were CT simulated and lungs were contoured using a treatment planning system. Anteroposterior/posteroanterior (AP/PA) fields were designed with MLC aperture conforming to lung contours. The fields were used to represent the extent of the lungs, which was subsequently marked on the patient's skin. The lung blocks were positioned with their shadow matching the lungs' marks. Their position was radiographically verified prior to the delivery of each beam. To evaluate the efficiency of this technique, the treatment session time and the number of repeated attempts to correctly position the shielding blocks was recorded for each beam. Exact treatment times for patients treated with the old technique were not available and were hence approximated based on previous experience. RESULTS: We succeeded in positioning the shielding blocks from the first attempt in 10/12 beams. The position of the shielding blocks was adjusted only one time prior to treatment in 2/12 beams. These results are compared to an average of 3 attempts per beam for each patient using the conventional technique of trial and error. The average time of a treatment session was 29 min with a maximum of 41 min versus approximately 60 min in past treatments and a maximum of 120 min. CONCLUSION: This new technique succeeded in reducing the length of the overall treatment session of the conventional TBI procedure and hence reduced patient discomfort while ensuring accurate shielding of the lungs.

Template-based breast IMRT planning for increased workload efficiency.

BACKGROUND: To be less resource intensive, we developed a template-based breast IMRT technique (TB-IMRT). This study aims to compare resources and dose distribution between TB-IMRT and conventional breast radiation (CBR). METHODS: Twenty patients with early stage breast cancer were planned using CBR and TB-IMRT. Time to plan, coverage of volumes, dose to critical structures and treatment times were evaluated for CBR and TB-IMRT. Two sided-paired t tests were used. RESULTS: TB-IMRT planning time was less than CBR (14.0 vs 39.0 min, p < 0.001). Fifteen patients with CBR needed 18 MV, and 11 of these were planned successfully with TB-IMRT using 6 MV. TB-IMRT provided better homogeneity index (0.096 vs 0.124, p < 0.001) and conformity index (0.68 vs 0.59, p = 0.003). Dose to critical structures were comparable between TB-IMRT and CBR, and treatment times were also similar (6.0 vs 7.8 min, p = 0.13). CONCLUSIONS: TB- IMRT provides reduction of planning time and minimizes the use of high energy beams, while providing similar treatment times and equal plans compared to CBR. This technique permits efficient use of resources with a low learning curve, and can be done with existing equipment and personnel.