syngo.via RT Image Suite

The syngo.via RT Image Suite is an advanced image analysis software designed to support radiation therapy planning by enabling comprehensive visualization, multi-modality manipulation, and precise segmentation of anatomical structures, including tumors and organs-at-risk. It leverages digital image processing techniques, incorporating AI-driven deep learning for functionalities such as automated lung lobe segmentation for ventilation assessment, to facilitate accurate contouring, image fusion, and the preparation of geometric treatment plans.

The syngo.via RT Image Suite, software version SOMARIS/8 VB70, is an image analysis software designed for radiation therapy planning. It enables viewing, manipulation, 3D/4D visualization, and comparison of medical images from multiple modalities (e.g., CT, PET, MRI, Linac CBCT). The software supports the segmentation of tumors and organs-at-risk through manual and software-assisted contouring tools, facilitates rigid and deformable image registration, and allows for the creation of simple geometric treatment plans which can be exported. Key modifications in this version include enhanced patient marking with simplified isocenter placement in the spine, a new tool for converting isodose lines to contours, improved contouring capabilities on 4D image data, the introduction of lobe-based lung ventilation, and an added software interface for advanced visualization and measurement tools. A significant feature of this software version is the AI-driven lobe-based lung ventilation calculation, which uses a deep learning algorithm for autosegmentation of lung lobes from inspiration and expiration CT images. The AI component, reused from a prior version, was trained on 8721 scans, validated on 969, and tested on an independent cohort of 18 radiotherapy patients, achieving a mean DICE coefficient of 0.92 for lung lobe segmentation. Clinical validation of the overall lung ventilation algorithm on 108 CT datasets from 74 patients demonstrated that ventilation distributions were comparable to literature-derived values in one subgroup (4D-CT, normal breathing). In another subgroup (breathhold CT, forced breathing), ventilation results showed a significant correlation (R=0.63, p<0.001) with spirometry-measured vital capacity. The document notes limitations, including reliance on literature for ground truth in one validation arm and potential discrepancies due to differing conditions between CT and spirometry tests, advising that the ventilation results should not be the sole diagnostic tool.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. Regarding the AI-based lung lobe segmentation, what was the variability in DICE scores across the 18 test patients, and were there specific anatomical variations or image qualities that correlated with lower scores?; How does the performance of the lobe-based lung ventilation algorithm, particularly its correlation with spirometry (R=0.63), vary in patient populations with different underlying pulmonary pathologies or varying degrees of tumor involvement in the lungs?; The 'simplified workflow to place isocenters in the spine' is mentioned; could you elaborate on the specific changes made and any quantitative data demonstrating improvements in efficiency or accuracy for this task?; For the tool that converts isodose lines from a DICOM RT dose file to contours, what is the validation process for this feature, and how does it ensure accuracy, especially in regions with steep dose gradients or complex isodose shapes?; Considering the stated limitations of the clinical validation for lung ventilation, such as reliance on literature for ground truth and differing exam conditions, what strategies could be employed in future studies to provide more direct, patient-specific validation of this feature's accuracy and clinical utility?