Table 2 The 15 steps needed to turn a fractal tool in a clinically valid test (also considering STARD [39] criteria)
From: Fractal frontiers in cardiovascular magnetic resonance: towards clinical implementation
Developmental step | Fractal quantification of trabecular complexity [9] | Fractal quantification of the spatial distribution of pulmonary flow [7] |
|---|---|---|
1. Technical development and theoretical basis of the test | Achieved – method first implemented in Java [8] and then in MATLAB [9] to improve computational efficiency; many segmentation algorithms tested before choosing a region-based level-set function [40] | Achieved – fractal dimension used as an index of pulmonary perfusion heterogeneity; image preparation included a coil inhomogeneity correction |
2. Comparison with gold-standard or tissue biopsy (animal models and then human biopsy material) | Achieved – validated against episcopic mouse embryo datasets and using synthetically constructed phantoms with well-known FD: 1) regular geometrical objects (plane, cube surface, sphere surface) and 2) ideal monofractal signals (4th, 5th and 6th iteration of the Sierpinski carpet or 9th, 10th and 11th iteration of the Sierpinski gasket) | Part achieved – validated using 3 MR reference phantoms applied to each patient’s chest |
3. Detection of changes in established disease compared with normals | Achieved – FD in left ventricular noncompaction compared to healthy volunteers | Not achieved |
4. Correlation with other equivalent cardiac imaging markers | Achieved – correlated with perimeter and with noncompacted/compacted wall thickness ratio [41] | Achieved – comparison is made with relative dispersion and the geometric standard deviation |
5. Correlation with other relevant biomarkers | Not achieved | Achieved – data correlated with pulmonary function test from spirometry and repeated for three different inspired oxygen concentrations (normoxia, hypoxia and hyperoxia) |
6. Demonstration of the test in more than one condition | Achieved – noncompaction and also subclinical and overt hypertrophic cardiomyopathy (and hypertension, in press) | Not achieved |
7. Demonstration of test sensitivity (early disease or change with age) | Achieved – in subclinical hypertrophic cardiomyopathy | Not achieved |
8. Demonstration of ability to track changes over time | Not achieved | Not achieved |
9. Demonstration of predictive or prognostic value of the test | Achieved – in combination with other CMR imaging markers, high FD shown to predict sarcomere gene mutation carriage in subclinical hypertrophic cardiomyopathy | Not achieved |
10. Standardization of the test (reproducibility, different equipment, in non-research settings, quality control, limitations of test) | Achieved – intra- and inter-observer variability, inter-scanner reproducibility, field-strength and slice-thickness independence demonstrated; community roll-out started through open-source release of an OsiriX plugin and development of an equivalent commercial version (in cvi42, Circle Cardiovascular Imaging) | Not achieved |
11. Development of robust age/ethnic normal reference ranges | Achieved – through analysis of the Multi-Ethnic Study of Atherosclerosis (in press); robust to multi-centre/multi-vendor implementation | Not achieved |
12. Changes in biomarker remain tied to the disease after treatment | Not achieved | Not achieved |
13. Demonstration of test as surrogate trial end point | Not achieved | Not achieved |
14. Clinical use and regulatory approval of test | Not achieved | Not achieved |
15. Prove that test use improves clinical outcome | Not achieved | Not achieved |