syngo.CT Lung CAD

The syngo.CT Lung CAD (VD20) is an advanced software device utilizing Convolutional Neural Networks to assist radiologists in the detection of solid and subsolid pulmonary nodules from multi-detector CT chest examinations. Its methodology encompasses image preprocessing with lung segmentation, CNN-driven nodule candidate generation and classification, and a novel CNN-based post-filtering module to enhance accuracy by reducing false positives, with studies confirming improved nodule detection rates when employed by radiologists.

The syngo.CT Lung CAD (VD20) is a software-based Computer-Aided Detection (CAD) tool designed to aid radiologists in identifying solid pulmonary nodules (between 3.0 mm and 30.0mm) and subsolid (part-solid and ground glass) nodules (between 5.0 mm and 30.0mm) in multi-detector computed tomography (MDCT) chest examinations. It functions as an adjunctive tool, processing images from multivendor CT scanners (16 or more detector rows) to highlight regions of interest (ROIs) that might be overlooked. The software can be integrated into clinical practice as either a concurrent first reader, assisting the radiologist during their initial review, or as a second reader after their primary assessment. Notably, the syngo.CT Lung CAD (VD20) is an algorithm that provides a list of nodule candidates to a separate hosting application, such as syngo MM Oncology, which is responsible for displaying the CAD marks on the images. The device's core technology relies on Convolutional Neural Networks (CNNs) for its main functions: preprocessing (including isotropic volume resampling and lung segmentation using a V-net CNN), candidate generation from subvolumes of the CT scan, and candidate classification to determine the likelihood of a detection being a nodule. Key enhancements in the VD20 version include new AI models for candidate generation and classification, regional configurability options, a CNN-based post-filtering module specifically aimed at reducing false positives caused by bony protrusions or detections in the colon, and bug fixes to improve lung segmentation, particularly in apical or basal lung regions. The effectiveness of these updates was supported by a retrospective reader study involving 20 readers and 232 cases, which demonstrated a significant improvement in pulmonary nodule detection by all readers using both reading workflows (concurrent and second reader) compared to unaided interpretation. Non-clinical testing further verified the software's performance and safety.

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. How does the performance of the CNN-based post-filtering module in reducing false positives (e.g., from bony protrusions or colon) quantitatively compare to the previous version (VC30) or to systems lacking this specific module?; What were the specific algorithmic changes or training data enhancements related to the bug fixes for apical or basal-lungs under-segmentation, and what was the measured impact on segmentation accuracy in these regions?; The reader study showed improvement with both concurrent first reader and second reader workflows. Was there a statistically significant difference in overall diagnostic accuracy, sensitivity, or radiologist reading time between these two workflows when using syngo.CT Lung CAD (VD20)?; Given the multivendor compatibility claim and the range of supported CT acquisition parameters (e.g., voltage, slice thickness, kernels), how robust is the syngo.CT Lung CAD (VD20)'s performance across these variations, particularly concerning detection sensitivity and false positive rates?; What are the standalone performance metrics (e.g., sensitivity per nodule, false positive rate per scan) of the syngo.CT Lung CAD (VD20) algorithm for detecting different types (solid, part-solid, ground glass) and sizes of pulmonary nodules, especially at the lower detection limits (3mm for solid, 5mm for subsolid)?