مقالات پذیرفته شده در نهمین کنگره بین المللی زیست پزشکی
Prostate Cancer Grading Using mpMRI-based Radiomics and Machine Learning
Prostate Cancer Grading Using mpMRI-based Radiomics and Machine Learning
Mahmoud Mohammadi-Sadr,1Farzaneh Kashi,2,*
1. Department of Medical Physics, School of Medicine, Isfahan University of Medical Sciences, Isfahan, Iran 2. Department of Medical Physics and Engineering, School of Medicine, Tehran University of Medical Sciences, Tehran, Iran
Introduction: Prostate cancer diagnosis still poses a major issue, requiring correct grading to determine treatment plans and optimize management expectations for patients. The Gleason grading approach is the most widely used method for evaluating prognosis of prostate cancer; however, routine histopathology can be disruptive and also have a degree of subjectivity. Recently developed radiomics and machine learning (ML) techniques have powerful potential in non-invasive grading of prostate cancer using multi-parametric magnetic resonance imaging (mpMRI) images that could offer comprehensive information about the tumor.
Methods: The present review is concerned with the latest studies that utilized radiomics and ML to classify the grade groups (GGs) levels of prostate cancer using mpMRI features. Classification included the extraction of the quantitative imaging features of mpMRI such as texture, shape, and intensity, which underwent analysis with multiple types of ML classifiers. Importantly, some model improvement techniques like Recursive Feature Elimination (RFE) were applied to remove irrelevant features and retain only those key to enhancing model classification performance.
Results: The findings of several investigations suggest that the radiomics based models can reach a high level of classification accuracy for distinguishing between different Gleason grades. For example, one study cited an astonishing 97% accuracy when high b-value diffusion weighted MRI features were used with RFE and Random Forest Classifier (RFC) for multiclass classification of prostate cancer into five GGs. In another study, a deep learning model was able to differentiate between clinically significant and indolent prostate cancers, achieving 87.2% accuracy for classifying lesions based on T2-weighted images. These studies highlight the possibility of fusion of radiomics and ML to improve accuracy of diagnosis in prostate cancer grading.
Conclusion: In conclusion, the application of radiomics features extracted from mpMRI images, coupled with machine learning techniques, represents a transformative approach to grading prostate cancer. This non-invasive methodology not only improves the accuracy of prostate cancer classification but also has the potential to reduce the number of unnecessary biopsies, thereby enhancing patient care. Future research should focus on validating these models in larger, diverse cohorts and exploring their integration into clinical practice to optimize prostate cancer management.