The present study validated measurements of complex anatomical features of the human heart by comparing IMVR and CPR with reference values. Both methods demonstrated excellent intra- and inter-observer reliability, with ICCs of ≥ 0.98, and negligible bias within 0.5 mm according to the BA analyses. However, IMVR showed better agreement with the reference values than CPR, as indicated by the LOA in the BA analysis; moreover, IMVR maintained consistent performance in both the aortic root and coronary artery measurements with a median speed of approximately 1 mm/s.
While some studies have validated the point location of a vascular model using a localization system of augmented reality [16] or linear measurements of a CT phantom using extended reality [17], this study focused on the measurement of more complex shapes of the heart. The measurement of various anatomical features using either IMVR or CPR requires sophisticated 3D interactions involving appropriate target orientation and point placement. In this experiment, we measured 3D-printed cardiac models duplicating the shapes of the components of the human heart. Physical measurements using a digital caliper or straight ruler were not possible for internal or recessed features of the models. Instead, reference values of the discrete model features were accurately obtained using a high-precision optical scanner.
When comparing the two techniques through BA analysis, the range of LOAs associated with IMVR was narrower than that of CPR. As shown in Table 2, the LOA of IMVR ranged within ± 3 mm in the aortic root and coronary artery measurements. In contrast, that of CPR was twice as wide (± 6 mm) in the aortic root measurement and thrice as wide (± 9 mm) in the coronary artery measurement. The BA plots in Figs. 5 and 6 show that the majority of values presented in IMVR showed little spread, independent of the object size. This suggests good accuracy in the measurement of relatively larger structures, such as crown-shaped aortic annuli, dilated sinuses of Valsalva, or saddle-shaped mitral annuli. Possible explanations for these better results are as follows: (1) visualized volume-rendered images representing the complex anatomical features of the cardiac models, including clearly identifiable radiopaque markers along the presumed path, allow visually corroborated 3D measurements; (2) superimposed surface-rendered images, representing precise isosurface geometry on volume-rendered images, effectively ensure the correct depth of placement of marker-points for measurement; and (3) IMVR require a fairly simple 3D operating technique to manipulate 3D images and place marker-points using a regular computer mouse, which can be easily learned by novice users.
Whether the absolute LOAs in IMVR are clinically acceptable will likely depend on the clinical requirements and, particularly, the overall dimensions of the structures of interest. For example, in technically demanding aortic valve-sparing or transcatheter valve implantation procedures, IMVR may provide clinically relevant preoperative information to surgeons. In such cases, a measurement error within 2 mm may be clinically acceptable for the selection of prosthetic valve or graft sizes [12, 18, 19]. Advanced 3D operating skills may be required for precise measurements to avoid misjudgment during procedures. Notably, with the experience gained in three trials, the measurement speed of the novice user (Observer #1) improved remarkably to match that of the experienced user (Observer #2) (see Supplementary Table S8), indicating the possible existence of a quick learning curve for 3D operation in IMVR. Additionally, measurement of the outer curvature length of the thoracic aorta is reportedly far more appropriate than centerlines in planning endovascular aneurysm repair [20]. Although the measurement accuracy of aortic lengths requires further investigation, IMVR would seem effectively applicable for the preoperative planning of thoracic endografting.
MPR and CPR are standard methods for most radiologists and cardiologists [2, 3, 21] and are available in most image-processing applications. In general, CPR depends on the accurate tracking of a curved path that can be represented within a single plane for review [21]. However, a significant disadvantage of these methods is the difficulty in correctly orienting anatomical features using multiplanar images, which influences marker-point placement in CPR applied to tortuous geometry. Throughout the trials, the CPR observers had difficulty tracing the presumed path along the anatomical features, which was consistent with the larger spread in the BA plots observed. Similarly, both CPR observers experienced remarkable difficulties in identifying coronary landmarks, especially those with tortuous shapes and small diameters. Indeed, the median speed of CPR measurements in the coronary artery was slower than that in the aortic root. As expected, 18 coronary segments that included “the tip” showed the lowest agreement in CPR (in Supplementary Table S4).
This study had some limitations. First, this study involved 3D-printed models composed of rigid resin and radiopaque markers for objective measurements and reference standards. Various tissue components might affect the measurement of a patient’s imaging data, including contrast-enhanced cardiac muscles or vessels with adjacent soft tissues and calcified plaques. The accuracy and reliability of IMVR in these living tissues should be investigated in future studies. Second, this study did not rigorously validate the quality of the volume-rendered images, which could have influenced the accuracy of IMVR. Apart from phantom models allowing verifiable objective measurements with reference values, the in vivo intracardiac anatomy remains obscure because image quality can be easily influenced by the thresholds of CT attenuation values [5, 22]. The displayed image can be further adjusted using transfer functions by altering the color and opacity levels or other parameters of the imaging algorithms [7, 22]. Therefore, several concerns have been raised regarding the direct interpretation of 3D images [5]. Before introducing IMVR in clinical applications, such rendering processes must undergo proper evaluation. The third limitation is the small number of observers. In this study, only two physicians were enrolled as examiners for each model. The user performance on IMVR, including that of a less-experienced operator, is worth considering in relation to the learning curve.
In conclusion, compared with conventional and fundamentally two-dimensional CPR, IMVR, which combines isosurface geometry with volume-rendered images, appears to be more precise, facilitating accurate 3D measurements of complex features of the heart. Based on better agreement and a quick learning curve associated with IMVR, this technique has the potential for use in myriad cardiovascular anatomies and clinical applications, including preoperative planning for complex valvular heart diseases or aortic aneurysms. Further research is warranted to evaluate the performance of IMVR in clinical images of patients.