Deep Learning Auto-Segmentation Network for Pediatric Computed Tomography Data Sets: Can We Extrapolate From Adults?

PURPOSE: Artificial intelligence (AI)-based auto-segmentation models hold promise for enhanced efficiency and consistency in organ contouring for adaptive radiation therapy and radiation therapy planning. However, their performance on pediatric computed tomography (CT) data and cross-scanner compatibility remain unclear. This study aimed to evaluate the performance of AI-based auto-segmentation models trained on adult CT data when applied to pediatric data sets and explore the improvement in performance gained by including pediatric training data. It also examined their ability to accurately segment CT data acquired from different scanners. METHODS AND MATERIALS: Using the nnU-Net framework, segmentation models were trained on data sets of adult, pediatric, and combined CT scans for 7 pelvic/thoracic organs. Each model was trained on 290 to 300 cases per category and organ. Training data sets included a combination of clinical data and several open repositories. The study incorporated a database of 459 pediatric (0-16 years) CT scans and 950 adults (>18 years), ensuring all scans had human expert ground-truth contours of the selected organs. Performance was evaluated based on Dice similarity coefficients (DSC) of the model-generated contours. RESULTS: AI models trained exclusively on adult data underperformed on pediatric data, especially for the 0 to 2 age group: mean DSC was below 0.5 for the bladder and spleen. The addition of pediatric training data demonstrated significant improvement for all age groups, achieving a mean DSC of above 0.85 for all organs in every age group. Larger organs like the liver and kidneys maintained consistent performance for all models across age groups. No significant difference emerged in the cross-scanner performance evaluation, suggesting robust cross-scanner generalization. CONCLUSIONS: For optimal segmentation across age groups, it is important to include pediatric data in the training of segmentation models. The successful cross-scanner generalization also supports the real-world clinical applicability of these AI models. This study emphasizes the significance of data set diversity in training robust AI systems for medical image interpretation tasks.

Evaluating synthetic computed tomography images for adaptive radiotherapy decision making in head and neck cancer.

Background and purpose: Adaptive radiotherapy (ART) decision-making benefits from dosimetric information to supplement image inspection when assessing the significance of anatomical changes. This study evaluated a dosimetry-based clinical decision workflow for ART utilizing deformable registration of the original planning computed tomography (CT) image to the daily Cone Beam CT (CBCT) to replace the need for a replan CT for dose estimation. Materials and methods: We used 12 retrospective Head & Neck patient cases having a ground truth - a replan CT (rCT) in response to anatomical changes apparent in the daily CBCT - to evaluate the accuracy of dosimetric assessment conducted on synthetic CTs (sCT) generated by deforming the original planning CT Hounsfield Units to the daily CBCT anatomy.The original plan was applied to the sCT and dosimetric accuracy of the sCT was assessed by analyzing plan objectives for targets and organs-at-risk compared to calculations on the ground-truth rCT. Three commercial DIR algorithms were compared. Results: For the best-performing algorithms, the majority of dose metrics calculated on the sCTs differed by less than 4 Gy (5.7% of 70 Gy prescription dose). An uncertainty of ±2.5 Gy (3.6% of 70 Gy prescription) is recommended as a conservative tolerance when evaluating dose metrics on sCTs for head and neck. Conclusions: Synthetic CTs present a valuable addition to the adaptive radiotherapy workflow, and synthetic CT dose estimates can be effectively used in addition to the current practice of visually inspecting the overlay of the planning CT and CBCT to assess the significance of anatomical change.

Selection of motion management in liver stereotactic body radiotherapy and its impact on treatment time.

Background and purpose: Reduction of respiratory tumour motion is important in liver stereotactic body radiation therapy (SBRT) to reduce side effects and improve tumour control probability. We have assessed the distribution of use of voluntary exhale breath hold (EBH), abdominal compression (AC), free breathing gating (gating) and free breathing (FB), and the impact of these on treatment time. Materials and Methods: We assessed all patients treated in a single institution with liver SBRT between September 2017 and September 2021. Data from pre-simulation motion management assessment using fluoroscopic assessment of liver dome position in repeat breath holds, and motion with and without AC, was reviewed to determine liver dome position consistency in EBH and the impact of AC on motion. Treatment time was assessed for all fractions as time from first image acquisition to last treatment beam off. Results: Of 136 patients treated with 145 courses of liver SBRT, 68 % were treated in EBH, 20 % with AC, 7 % in gating and 5 % in FB. AC resulted in motion reduction < 1 mm in 9/26 patients assessed. Median treatment time was higher using EBH (39 min) or gating (42 min) compared with AC (30 min) or FB (24 min) treatments. Conclusions: Motion management in liver SBRT needs to be assessed per-patient to ensure appropriate techniques are applied. Motion management significantly impacts treatment time therefore patient comfort must also be taken into account when selecting the technique for each patient.