Review of physics-based models for coronary stenotic pressure losses: a comparative study with in vitro experiments and clinical data

Abstract Background The Fractional Flow Reserve (FFR) is the gold standard for functional assessment of coronary stenosis severity through invasive measurements (FFRinv). While coronary angiography-derived physiology (FFRangio) is gaining clinical acceptance, a recent core lab comparison of commercialized software showed moderate performance, with area under the receiver-operating characteristic curve (AUC) ranging from 0.73 to 0.75 (1). These non-invasive methods often use three-dimensional (3D) vessel reconstructions and physics-based models describing the stenotic pressure loss using complex 3D simulations to simpler stenosis models. Purpose This study aims to: i) review literature for physics-based models of coronary stenosis, and ii) evaluate their performance in computing stenotic pressure losses using in vitro experiments and clinical data. Methods A literature review identified original zero-dimensional physics-based models for stenotic pressure loss. These models were implemented in Matlab (The MathWorks Inc.) and validated with in-vitro pressure losses in 12 stenosis geometries, either block- or gradient-shaped, using water as fluid. These had a reference diameter of 3.5mm, with diameter severity (DS) ranging from 50 to 80% and lesion lengths (Ls) between 4–40 mm. Experimental flow rates varied between 50 to 200 mL/min. Moreover, lesions from the FAME study (2)were analysed using angiogram-derived lesion characteristics and estimations of coronary flow as model inputs. Serial lesions were excluded from analyses. Results A total of 27 models was successfully extracted and implemented. In-vitro experiments showed that most models underestimated the coronary pressure loss compared to actual measurements, with greater variation in model prediction at higher flows (Fig.1). There was no single model that provided accurate outcomes for all variations of geometries and flow rates. A total of 600 lesions from the FAME study was included, with an average reference diameter of 2.5 ± 0.6 mm, a mean diameter severity of 59.2 ± 16.2%, and mean length of 12.8 ± 5.9 mm. The mean FFRinv was 0.72 ± 0.17, and the flow was estimated at 110 ± 59 mL/min. Fourteen models were applicable to the available clinical data. The AUC values of these models ranged from 0.71 to 0.75, with one outlier (AUC = 0.43) (Fig.2). Conclusion No single physics-based model was found to be universally optimal for the various conditions in the in vitro experiments, with large variations in predicted pressure losses. When applied to clinical data, the models showed AUCs similar to commercially available software. This suggests that the selection of a model for FFRangio computation should be made with careful consideration of the specific clinical scenario and could be based on simple stenosis characteristics instead of 3D reconstructed geometries.Model versus in-vitro pressure losses ROC curve for all models