Enhancing Prostate Cancer (PCa) Diagnosis: Precision Analysis of Diagnostic Accuracy (DA) using 68Ga-PSMA PET/CT based GLCM Texture Parameters

Introduction: The prostate Specific Membrane Antigen (PSMA), a type II transmembrane glycoprotein of the prostate secretory acinar epithelium, is up-regulated in prostate carcinoma and its metastases. ⁶⁸Ga-PSMA, a PET tracer recently introduced for the imaging of patients with prostate cancer (PCa) at diagnosis, staging/restaging and to assess the patients with any post-treatment biochemical failure. However, PET/CT has limitation to characterize subtle tissue variation and tissue or tumor heterogeneity. PCa patients are typically diagnosed and staged using Transrectal Ultrasonography (TRUS) guided biopsy. The Gleason score is frequently used to estimate the aggressiveness of PCa lesions. Recently, literature suggested texture parameter may further improve the diagnostic accuracy of PCa and may provide a non-invasive tool unlike biopsy for the grading and staging, by analyzing tumor heterogeneity and complexity of the lesions. This study explores improved PCa diagnosis by combining ⁶⁸Ga-PSMA PET/CT with radiomics features, specifically focusing on Gray-Level Co-occurrence Matrix (GLCM) texture parameters to enhance diagnostic accuracy.

Methods: We ambispectively reviewed PCa patients who underwent ⁶⁸Ga-PSMA PET/CT to know the disease status. Patients were categorized into healthy and malignant PCa patients based on PSMA expression on ⁶⁸Ga-PSMA PET/CT. Gray-Level Co-occurrence Matrix (GLCM) Haralick texture features were obtained as a result of texture analysis performed using MATLAB (v. 2018; MathWorks, Natick, MA, USA). Normality of the data was checked with Shapiro-Wilk test. Accordingly, Man Whitney U-test tests were applied to check the significant difference (p>0.05) in the mean of texture parameters between normal and malignant prostate cancer patients. Receiver Operative Characteristic (ROC) curve analysis was done on significant parameters to estimate the cut-off values along with corresponding sensitivity and specificity. All the analysis was done using SPSS software v22.0.

Results: A total of 150 biopsy proven healthy and malignant PCa patients with mean age of 68.18±7.08 years were included in this study. However, final analysis was performed on 119 patients (37 normal; 82 malignant PCa patients) and due to non-availability of Gleason score. All 9 GLCM features were found to be statistically significant. The concordance and discordance of texture features was determined based on the Gleason score as reference standard. The sensitivity, specificity, diagnostic accuracy (DA), positive predictive value (PPV) of GLCM texture parameters such as energy, contrast, entropy, homogeneity, correlation, sum average, variance, dissimilarity, autocorrelation was found to be [93.18%; 100%; 100%; 94.96%], [92.05%; 96.77%; 98.78%; 93.28%], [93.18; 100%; 100%; 94.96%], [93.18%; 90.32%; 96.47%; 92.44%], [85.23%; 70.97%; 89.29%; 81.51%], [93.18%; 100%; 100%; 94.96%], [93.18%; 96.77%; 98.8%; 94.12%], [93.18%; 93.55%; 97.62%; 93.28%], [93.18%; 100%; 100%; 94.96%]. Among all GLCM texture parameters, energy, entropy, sum average and autocorrelation showed the highest DA of 94.96%.

Conclusions: The GLCM texture parameters particularly, energy, entropy, sum average and autocorrelation showed the highest diagnostic accuracy to differentiate between healthy and malignant PCa patients.