Validation and diagnostic accuracy of Fractal Dimension as a biomarker for grading prostate cancer using data from clinical routine.

OBJECTIVES: To evaluate fractal dimension (FD) as a quantitative prostate cancer (PCa) biomarker for the International Society of Urological Pathology Grade Group (ISUP-GG) and validate its diagnostic accuracy using clinical data from multi-scanner sources. METHODS: Over a 20-month period, 153 men with biopsy-proven PCa aged 48.6-83 years (mean 67.0±7.1 years; ISUP-GG 1-5 distribution: n = 2, 68, 44, 21 and 18), imaged on multiple scanners with differing fat-saturation protocols, had FD retrospectively calculated from the edge of the index lesion. Signal Intensities (SI) on dynamic contrast-enhanced magnetic resonance images (DCE-MRI) were normalised to the obturator internus muscle. For reproducibility, 20 patients had repeat FD extracted by an independent observer. FD was correlated with ISUP-GG. Previously determined thresholds were applied to determine diagnostic accuracy of FD for separating ISUP-GGs. RESULTS: After image normalisation, tumour SI values showed similar distributions between scanners. There was a significant correlation between FD and ISUP-GG (R = 0.87, p < 0.000001). A pre-specified threshold from previous datasets of 2.31 separated ISUP-GG1-2 from 3-5 with 95.2% sensitivity, 92.9% specificity and 94.1% diagnostic accuracy. In 20 randomly selected cases, mean inter-reader FD difference was -0.01, Bland-Altman limits of agreement -0.08 to + 0.05. CONCLUSIONS: FD is an accurate biomarker for grading PCa using standard clinical DCE-MRI with differing fat-saturation schemes that affect SI. ADVANCES IN KNOWLEDGE: Fractal dimension of DCE-MRI allows for non-invasive stratification of PCa and generalises to different MRI scanners and fat-saturation protocols.