Affiliation:
1. Department of Radio‐Diagnosis and Interventional Radiology All India Institute of Medical Sciences New Delhi India
2. Centre for Biomedical Engineering Indian Institute of Technology Delhi New Delhi India
3. Department of Nuclear Magnetic Resonance All India Institute of Medical Sciences New Delhi India
4. Department of Biostatistics All India Institute of Medical Sciences New Delhi India
5. Department of Urology All India Institute of Medical Sciences New Delhi India
6. Department of Pathology All India Institute of Medical Sciences New Delhi India
7. Department of Gastroenterology (Molecular Biology Division) All India Institute of Medical Sciences New Delhi India
8. Department of Nuclear Medicine All India Institute of Medical Sciences New Delhi India
Abstract
ObjectivesTo evaluate the role of combined intravoxel incoherent motion and diffusion kurtosis imaging (IVIM–DKI) and their machine‐learning‐based texture analysis for the detection and assessment of severity in prostate cancer (PCa).Materials and methodsEighty‐eight patients underwent MRI on a 3 T scanner after giving informed consent. IVIM–DKI data were acquired using 13 b values (0–2000 s/mm2) and analyzed using the IVIM–DKI model with the total variation (TV) method. PCa patients were categorized into two groups: clinically insignificant prostate cancer (CISPCa) (Gleason grade ≤ 6) and clinically significant prostate cancer (CSPCa) (Gleason grade ≥ 7). One‐way analysis‐of‐variance, t test, and receiver operating characteristic analysis was performed to measure the discriminative ability to detect PCa using IVIM–DKI parameters. A chi‐square test was used to select important texture features of apparent diffusion coefficient (ADC) and IVIM–DKI parameters. These selected texture features were used in an artificial neural network for PCa detection.ResultsADC and diffusion coefficient (D) were significantly lower (p < 0.001), and kurtosis (k) was significantly higher (p < 0.001), in PCa as compared with benign prostatic hyperplasia (BPH) and normal peripheral zone (PZ). ADC, D, and k showed high areas under the curves (AUCs) of 0.92, 0.89, and 0.88, respectively, in PCa detection. ADC and D were significantly lower (p < 0.05) as compared with CISPCa versus CSPCa. D for detecting CSPCa was high, with an AUC of 0.63. A negative correlation of ADC and D with GS (ADC, ρ = −0.33; D, ρ = −0.35, p < 0.05) and a positive correlation of k with GS (ρ = 0.22, p < 0.05) were observed. Combined IVIM–DKI texture showed high AUC of 0.83 for classification of PCa, BPH, and normal PZ.ConclusionD, f, and k computed using the IVIM–DKI model with the TV method were able to differentiate PCa from BPH and normal PZ. Texture features of combined IVIM–DKI parameters showed high accuracy and AUC in PCa detection.