Xingfeng Li*
Department of Surgery & Cancer, Imperial College, London, UK
*Corresponding author:Xingfeng Li, Department of Surgery & Cancer, Imperial College, London, UK
Submission: April 19, 2023;Published: May 02, 2023
ISSN: 2640-9666Volume5 Issue4
Endometrial cancer is the second most prevalent cancer among women after breast cancer with an estimated 65,950 new cases diagnosed in 2022 in the USA alone [1]. To diagnose and make a treatment plan for endometrial cancer patients, MRI scans were often employed to assess the metastasis, Depth of Myometrial Invasion (DMI), Lymph Vascular Space Invasion (LVSI), and cancer risk score with an expert. However, the accuracy of the assessment results was highly dependent on the experience and knowledge of the experts, and therefore these results were subjective. To overcome this limitation, Machine Learning (ML) methods combined with radiomics attract considerable attention for endometrial cancer studies as it is automatic and objective for the patient diagnosis. Radiomics, which is an emerging method for identifying features from medical images [2-4], have been applied for endometrial cancer classification [5,6] and survival analysis studies [7]. The current methods, results, and future directions for endometrial cancer studies using ML and MRI radiomics were presented in the following.
For endometrial cancer classification studies, DMI, LVSI, and risk score were often adopted as response variables in the ML models, while the covariates were clinical variables and/or radiomic features. Studies have shown that these response variables can be predicted with radiomic features from multisequence MRI including T1-weighted, T2-weighted, and Diffusion Weighted Imaging (DWI). Based on the radiomic features obtained from MRI, the accuracy of the endometrial cancer response variable classification accuracy is different [8]. For example, for the LVSI prediction, the Area under the Receiver Operating Characteristic curve (AUC) of 0.80 has been achieved [9]. For the DMI classification, the AUC of 0.81 was achieved using multisequence MRI [9]. However, there was smaller AUC value for the risk score stratification using radiomic features. For instance, a recent study showed that the AUC of 0.72 was obtained from the T2-weighted MRI [10] for endometrial cancer patient risk classification. For most of the studies, the risk response variable classified the patient groups into two groups only, i.e., high and low risk groups. Future studies should investigate classification methods for 4 risk groups, i.e., low, intermediate, high, and advanced endometrial cancer patient groups [6]. For the endometrial cancer histology type classification using radiomic features and clinical features, high classification results (AUC=0.90) were achieved [6].
Compared with classification studies, fewer studies focused on using survival analysis methods to investigate the survival time of the endometrial cancer patient. Survival analysis is a collection of statistical procedures for data analysis where the outcome variable of interest is the time until an event such as death or recurrence occurs [11]. For example, the 5 year overall survival rate ranges from 74% to 91% in endometrial cancer patients without metastatic disease [12], but the 5-year overall survival rate for all cancer stage IIIA patients was 55% [13]. Estimating survival time is important as it is useful for patient management and making a treatment plan. There are three methods to study survival analysis, namely, non-parametric methods such as Kaplan Meier (KM) estimate [14], semi-parametric methods such as Cox Proportional Hazard (CPH), and parametric methods. KM and CPH model methods are the most popular non-parametric and semiparametric methods which have been carried out for endometrial survival analysis [15]. An integrated model combining clinical and radiomic features for the survival analysis has been developed [7]. In the analysis, radiomic features were incorporated into the CPH model for the survival time estimation. Although a deep learning survival analysis method has been developed [16], there is no report to use this method with radiomic features for endometrial cancer survival studies [17].
In summary, with the development of ML methods and radiomics, it is possible to classify endometrial cancer patients with a reasonable accuracy (e.g., AUC>0.8) using clinical and radiomic features. The majority of current studies adopted a handcrafted radiomics method, which requires manual segmentation of the tumour mask from medical images, while the deep learning based method is fully automatic to classify the patient into different groups for treatment [18], which is one of the future directions for endometrial cancer classification and survival analysis studies. More work needs to be done to further explore these methods and to improve the classification and survival time estimation accuracy.
© 2023 Xingfeng Li. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and build upon your work non-commercially.