Automatic Segmentation Using Deep Learning to Enable Online Dose Optimization During Adaptive Radiation Therapy of Cervical Cancer.

PURPOSE: This study investigated deep learning models for automatic segmentation to support the development of daily online dose optimization strategies, eliminating the need for internal target volume expansions and thereby reducing toxicity events of intensity modulated radiation therapy for cervical cancer. METHODS AND MATERIALS: The cervix-uterus, vagina, parametrium, bladder, rectum, sigmoid, femoral heads, kidneys, spinal cord, and bowel bag were delineated on 408 computed tomography (CT) scans from patients treated at MD Anderson Cancer Center (n = 214), Polyclinique Bordeaux Nord Aquitaine (n = 30), and enrolled in a Medical Image Computing & Computer Assisted Intervention challenge (n = 3). The data were divided into 255 training, 61 validation, 62 internal test, and 30 external test CT scans. Two models were investigated: the 2-dimensional (2D) DeepLabV3+ (Google) and 3-dimensional (3D) Unet in RayStation (RaySearch Laboratories). Three intensity modulated radiation therapy plans were generated on each CT of the internal and external test sets using either the manual, 2D model, or 3D model segmentations. The dose constraints followed the External beam radiochemotherapy and MRI based adaptive BRAchytherapy in locally advanced CErvical cancer (EMBRACE) II protocol, with reduced margins of 5 and 3 mm for the target and nodal planning target volume. Geometric discrepancies between the manual and predicted contours were assessed using the Dice similarity coefficient (DSC), distance-to-agreement, and Hausdorff distance. Dosimetric discrepancies between the manual and model doses were assessed using clinical indices on the manual contours and the gamma index. Interobserver variability was assessed for the cervix-uterus, parametrium, and vagina for the definition of the primary clinical target volume (CTVT) on the external test set. RESULTS: Average DSCs across all organs were 0.67 to 0.96, 0.71 to 0.97, and 0.42 to 0.92 for the 2D model and 0.66 to 0.96, 0.70 to 0.97, and 0.37 to 0.93 for the 3D model on the validation, internal, and external test sets. Average DSCs of the CTVT were 0.88 and 0.81 for the 2D model and 0.87 and 0.82 for the 3D model on the internal and external test sets. Interobserver variability of the CTVT corresponded to a mean (range) DSC of 0.85 (0.77-0.90) on the external test set. On the internal test set, the doses from the 2D and 3D model contours provided a CTVT V42.75 Gy >98% for 98% and 91% of the CT scans, respectively. On the external test set, these percentages were increased to 100% and 93% for the 2D and 3D models, respectively. CONCLUSIONS: The investigated models provided auto-segmentation of the cervix anatomy with similar performances on 2 institutional data sets and reasonable dosimetric accuracies using small planning target volume margins, paving the way to automatic online dose optimization for advanced adaptive radiation therapy strategies.

Evaluation of respiratory-induced target motion for esophageal tumors at the gastroesophageal junction.

PURPOSE: To quantify the internal motion margin requirements for radiotherapy of tumors near the gastroesophageal junction (GEJ). METHODS AND MATERIALS: Four-dimensional computed tomography (4DCT) scans were obtained for 25 patients with primary tumors located near the GEJ. The gross tumor volume (GTV) was manually contoured on the exhale-phase image from the 4DCT image set. A deformable image registration method was used to automatically propagate the contours to other phases of the 4DCT images. To quantify target motion, we measured the displacement of the GTV centroid and the variations in the target boundary and volume. Internal margins were calculated in the lateral (RL), anterior-posterior (AP), and superior-inferior (SI) directions. RESULTS: The mean+/-standard deviation peak-to-peak GTV centroid motion was 0.39+/-0.27cm (range, 0.04-1.09cm) in the RL, 0.38+/-0.23cm (range, 0.10-0.94cm) in the AP, and 0.87+/-0.47cm (range, 0.43-2.63cm) in the SI directions, respectively. On average, the internal target volume was 72% (range, 9-172%) larger than the GTV defined on a single-phase CT image. Variations in tumor boundaries due to tissue motion and deformation suggested asymmetric margins: 1.0cm left [toward the stomach], 0.8cm right, 1.1cm anterior, 0.6cm posterior, 1.0cm superior (toward the distal esophagus), and 1.6cm inferior (toward the stomach). CONCLUSION: Because tumors near the GEJ are subject to a marked but asymmetric amount of respiratory-induced intrafractional tumor motion, the use of asymmetric internal margins may be beneficial.

Daily alignment results of in-room computed tomography-guided stereotactic body radiation therapy for lung cancer.

PURPOSE: To determine the extent of interfractional setup errors and day-to-day organ motion errors by assessing daily bone alignment results and changes in soft tissue tumor position during hypofractionated, in-room computed tomography (CT)-guided stereotactic body radiation therapy (SBRT) of lung cancer. METHODS AND MATERIALS: Daily alignment results during SBRT were analyzed for 117 tumors in 112 patients. Patients received 40-50 Gy of SBRT in four to five fractions using an integrated CT-LINAC system. The free-breathing CT scans acquired during treatment setup were retrospectively realigned to match with each of the bony references and the gross tumor volume (GTV) defined on the reference CT by rigid-body registration, and the daily deviations were calculated. RESULTS: The mean magnitude (± SD) three-dimensional shift from the initial skin marks to the final bone-aligned positions was 9.4 ± 5.7 mm. The mean daily GTV deviation from the bone position was 0.1 ± 3.8 mm in the anterior-posterior direction, -0.01 ± 4.2 mm in the superior-inferior direction, and 0.2 ± 2.5 mm in the lateral direction. A clinically noteworthy trend (net change >5 mm in any direction) in GTV position relative to the bone was observed in 23 cases (20%). CONCLUSIONS: Soft tissue target position can change significantly beyond the motion envelope defined in the original internal target volume in four-dimensional CT-based treatment planning for SBRT of lung cancer. Additional margin should be considered for adequate coverage of interfractional changes.

Effectiveness of using fewer implanted fiducial markers for prostate target alignment.

PURPOSE: To evaluate the impact of the number and location of intraprostatic fiducial markers on the accuracy and reproducibility of daily prostate target alignment and to evaluate the migration of such markers. METHODS AND MATERIALS: Three gold fiducial markers were implanted transrectally under ultrasound guidance near the apex, middle, and base of the prostate in 10 prostate cancer patients. The patients had pretreatment in-room computed tomography (CT) scans three times a week, for approximately 25 CT scans per patient during the 8-week treatment course. A total of 1280 alignments were performed using different alignment scenarios: whole-prostate soft tissue alignment (the gold standard), bone alignment, and seven permutations of alignments using one, two, or three fiducial markers. The results of bone alignment and fiducial alignment were compared with the results of whole-prostate alignment. Fiducial migration was also evaluated. RESULTS: Single-fiducial-marker alignment was more accurate and reproducible than bone alignment. However, due to organ deformation, single fiducial markers did not always reliably represent the position of the entire prostate. The use of two-fiducial combinations was more accurate and reproducible than single-fiducial alignment, and use of all three fiducials was the best. Use of an apex fiducial together with a base fiducial rivaled the use of all three fiducials markers together. Fiducial migration was minimal. CONCLUSIONS: The number and the location of implanted fiducial markers affect the accuracy and reliability of daily prostate target alignment. The use of two or more fiducial markers is recommended.