Feature Extraction and Analysis of Prostate Cancer MR Images

Feature Extraction and Analysis based on Radiomics, which is a quantitative extraction and minable data analysis from radiology images, and it is establishing itself as one of the potential methods for the prostate lesion detection and classification. Prostate Cancer (PCa) prediction is a difficult task as it lacks of a solid biomarker and heterogeneity within and between Prostate Gland. As a result, quantitative analysis of imaging data adopting radiomics approach has developed into a new paradigm. This approaches in PCa allowed automated localization of the diseases, but it has also provided a solution that does not require the use of intrusive techniques to evaluate the biology of tumors. The effort is to assess the limitations and possibilities of Radiomics on magnetic resonance imaging (MRI) for predicting the treatment and long-term outcomes in PCa. Radiomics analyses were performed on MRI with 200 images of 60 patients. A review of literature of Radiomics reported a Heterogeneous results regarding Radiomics techniques and features included, but it however looks promising for the outcome prediction. Radiomics PCa research in future should concentrate on individual validation of conventional methods rather than on designing the new methods. The analysis of twenty-nine Features indicated that the original_firstorder_Kurtosis has the highest Importance as a feature in detection of Prostate Cancer with a value of 0.07100493960856451 followed by original_ngtdm_Coarseness as a second highest Importance value of 0.06048193882341631 followed by original_ngtdm_Busyness as a third highest Importance value of 0.0501837801054728. The proposed model performance recorded with the precision score was 0.84, the sensitivity or recall score was 0.86, and f1-score was 0.85.