Prostate cancer (Pca) is the most common malignancy and the second most dominant cause of cancer-related deaths in men in Western countries, with an incidence rate composed of 200 per 100,000 men.1 The early diagnosis of Pca will lead to an obvious increase in patients’ survival, and as a result, a decrease in treatment costs.2 Rapid and accurate diagnosis of Pca is crucial for improving outcomes.
Multiparametric prostate MRI (mp-MRI) is an emerging medical imaging modality that combines anatomic MRI with functional MRI technique, allowing for diagnosis, characterization, staging, and treatment strategy of Pca.3 T2-weighted imaging (T2WI) provides abundant anatomical details and makes major contributions to the localization and characterization of abnormalities in the prostate.4 T2WI images are capable of identifying the tumor lesion and assessing seminal vesicles, neurovascular bundles as well as prostate margins since T2WI allows for depicting the zonal anatomy and capsules, while the peripheral zone (PZ) is typically of intermediate to high signal intensity (SI) on T2WI images.5,6 As it is reported, there exists a strong correlation between tumor density and various tumor biological markers including Gleason score, tumor stage, as well as surgical margin status.7 Therefore, many studies have investigated the potential value of T2WI in localization of Pca. Recently, precision medicine has been introduced into routine clinical care with an increasing number of treatments being tailored to patient-specific characteristics. As a result, radiomics-based quantitative analysis for imaging data has been popularly and widely utilized.8
A recently published article in Exploratory Research and Hypothesis in Medicine by Ng et al. has compared different kernels of support vector machines (SVMs), one of the most popular supervised learning algorithms, to classify prostate cancerous tissues.7 The application of MRI using SVM algorithm with different kernels has not only enabled automatic classification of prostate cancerous tissue but also provided a non-invasive solution to assess Pca. The outlined merits could be summarized as follows: 1) Different SVM Kernels have been used to classify Pca; 2) Pca patients have been recruited from their own hospital instead of the public databases. In their study, machine learning is adopted since computer-aided detection and diagnosis calculated by machine learning facilitates interpreting medical imaging findings and reducing interpretation times. The article by Ng et al. is a very thorough, beautifully written and illustrated research paper on the utilization of different kernels in SVMs to classify prostate cancerous tissues on T2WI images as compared to previous studies, whose limitations could be attributed to the exclusion of late stages of Pca,9 without reference to the kernel used for the SVMs algorithm,10 or a lower yield accuracy.11 In conclusion, 17 features are extracted from the demarcated region of interest (ROI), and 5 features are retained by the principle component analysis (PCA) for SVM classification with the utilization of radical basis function (RBF), Gaussian, and lineal kernels. Consequently, SVMs using RBF yield the largest sensitivity and the second-largest accuracy.
The proposed application of MRI using SVM algorithm with different kernels could pave the way for identification of prostate cancerous tissue in a non-invasive method. In future, mpMRI combining anatomical MR imaging with functional MRI sequences could provide more useful information in Pca detection, disease monitor during active surveillance and patient follow-up.
Abbreviations
- mp-MRI:
multiparametric prostate MRI
- Pca:
prostate cancer
- RBF:
radical basis function
- SVMs:
support vector machine
- T2WI:
T2-weighted imaging
Declarations
Funding
This study was funded by the National Natural Science Foundation of China (Grant No. 82202135) for Shuai Ren and the Innovative Development Foundation of Department in Jiangsu Hospital of Chinese Medicine (Grant No. Y2021CX19) for Shuai Ren.
Conflict of interest
The authors have no conflicts of interest related to this publication.
Authors’ contributions
Drafting of the manuscript (SR, JW, and ZW), critical revision of the manuscript (SR), supervision (ZW).