DIVAID software validation shows high accuracy for atrial geometry annotation compared to expert manual methods.
This review and software development report evaluates the DIVAID software, an open-source tool designed for the automatic division of bi-atrial geometries. The validation study analyzed 140 geometries sourced from multiple modalities and centers, comparing DIVAID outputs against manual expert annotations. The primary outcome assessed the accuracy of regional boundaries using the Dice similarity coefficient and mean average surface distance, while secondary outcomes included vein clipping accuracy, orifice annotation accuracy, and speed and consistency.
The software demonstrated an orifice annotation accuracy of 100% and a Dice similarity coefficient of 0.98 for left atrial regions when compared to experts. For right atrial regions, the Dice similarity coefficient was 0.90 for DIVAID versus experts and 0.88 for inter-expert comparisons. Mean average surface distance for the left atrium was 0.17 mm, whereas the right atrium showed a larger distance of 1.93 mm. Vein clipping accuracy was reported at 81%.
The authors state that DIVAID enables large-scale, standardized regional analyses and data-driven investigation of harmonized, multi-dimensional datasets. This capability may advance atrial arrhythmia research and personalized treatment strategies. However, the report does not report adverse events, discontinuations, or tolerability. The authors caution that these validation metrics should not be interpreted as evidence of clinical efficacy or improved patient outcomes, as no clinical trial data is presented.