| Objective : To investigate the value of transrectal ultrasound combined with support vector machine(SVM)algorithm for the construction of radiomics model in the diagnosis of prostate cancer.Methods : Patients who underwent transrectal ultrasound-guided prostate puncture biopsy at the First Affiliated Hospital of Dalian Medical University from August 2021 to November 2022 were retrospectively collected.Transrectal 2D ultrasound was completed before biopsy,and those who had undergone prostate surgery,radiotherapy or had a history of other malignant tumours before 2D ultrasound were excluded.62 patients were collected with pathological findings as the gold standard,including There were 28 cases of prostate cancer and 34 cases of benign prostate disease;PSA was in the range of 4-10ng/ml in 32 patients,including 12 cases of prostate cancer and 20 cases of benign prostate disease.The transrectal 2D ultrasound images were acquired,uploaded to the Meditrans Darwin Intelligent Imaging Cloud Platform,and the Mask outlining tool of the platform was applied to manually outline along the contour of the lesion to obtain the region of interest(ROI),extract the imaging histology features in the region of interest,and perform the minimum-maximum normalization preprocessing,optimal feature screening and minimum-redundancy maximum The platform randomly divided the 62 patients into 49 patients in the training set and 13 patients in the test set in a ratio of 8:2;the 32 patients with PSA values between 4 and 10 ng/ml were randomly divided into 25 patients in the training set and 7 patients in the test set.Darwin used the subject operating characteristic(ROC)curve and area under the curve(AUC)to evaluate the diagnostic performance of the support vector machine classifier algorithm for prostate cancer and to calculate its accuracy,sensitivity and specificity.The comparison of patient clinical information was statistically analysed using IBM SPSS 26.0 software.Results : 1.General data results: the mean age of patients with prostate cancer was 77.64 ± 6.88 years(range 59-87 years)and the mean serum PSA was 68.13±102.07ng/ml(7.68-309.3ng/ml);the mean age of patients with benign prostate lesions was 65 ± 7.00 years(range 55-77 years)and the mean serum PSA was 11.73 ± 6.44ng/ml(1.88-27.28ng/ml),with statistically significant differences(P <0.05).2.Transrectal 2D ultrasound diagnostic results: the AUC value was 0.71,with 69% accuracy,78% sensitivity and 62% specificity.3.Imaging histology feature extraction and screening results: a total of 1125 features were extracted from the ultrasound images in this study features,and a total of 10 features were screened after two components of optimal feature screening and minimum redundancy maximum correlation,among which texture features accounted for five,namely normalized inverse disparity of GLCM,grey level inhomogeneity of GLDM,regional variance of GLSZM and long travel low grey focus and short range focus of GLRLM,and the other five were first-order features,namely 10 th percentile,uniformity,skewness,minimum and median;with the best feature being the normalised inverse disparity of the GLCM.4.Diagnostic results of imaging histology for prostate cancer: The AUC of the test group of the model was: 0.83,with an accuracy of 76.9%,sensitivity of 71.4% and specificity of 83.3%,and its AUC value was higher than that of the transrectal 2D ultrasound diagnosis.The AUC of the model test group was 0.80 when the PSA value was between 4 and 10 ng/ml,with an accuracy of 57.1%,sensitivity of 60% and specificity of 50%,which also had a better diagnostic performance.Conclusion : The extraction of histological features of the prostate based on transrectal ultrasound images,combined with the support vector machine(SVM)algorithm to construct an radiomics model,has good predictive performance for prostate cancer and can provide an objective,accurate,non-invasive and convenient method for the diagnosis of prostate cancer. |
PDF Full Text Request