Tehran University Medical Journal

Background: Multiple sclerosis (MS) is a degenerative inflammatory disease which is most commonly diagnosed by magnetic resonance imaging (MRI). But, since the MRI device uses of a magnetic field, if there are metal objects in the patient's body, it can disrupt the health of the patient, the functioning of the MRI, and distortion in the images. Due to limitations of using MRI device, screening seems necessary for those patients who have metal objects in their bodies. Therefore, this study is carried out to compare two models: support vector machine and random forest.
Methods: This analytical-modelling research was implemented on MS data collection, the specifications of which are recorded in health registry system in School of Public Health, Kermanshah University of Medical Sciences, Iran, from May 2017 to August 2018. For the purpose of this study, a total of 317 subjects were selected as a sample; 188 subjects were diagnosed with MS and 128 subjects showed no symptoms of MS. In order to fit the support vector machine (SVM) model, radial basis kernel function was used. The parameters of this machine were optimized with genetic algorithm. After this step, the support vector machine and random forest (RF) were compared with respect to three factors: accuracy, sensitivity, and specificity.
Results: Based upon the obtained results of study, accuracy, sensitivity, and specificity of SVM were 0.79, 0.80, and 0.78, respectively. In comparison, accuracy, sensitivity, and specificity of RF were found to be 0.76, 0.81, and 0.70, respectively.
Conclusion: In general, both models which were compared in current study showed desirable performance; however, in term of accuracy, as an important criteria for performance comparison in this area of research, it can be argued that support vector machine can do better than random forest in diagnosing multiple sclerosis.

Background: In many studies, Cox regression was used to assess the important factors that affect the survival of cancer patients based on demographic and clinical variables. The aim of this study was to determine the factors affecting the survival of patients with Hodgkin's lymphoma using the random survival forest (RSF) method and compare it with the Cox model.
Methods: In this retrospective cohort study, all patients with Hodgkin's lymphoma who were referred to the Oncology and Hematology Center of Ahvaz Shafa Hospital from March 2000 to February 2010 were included. The survival time was calculated from diagnosis to the first recurrence event date (based on month). To assess the process of the disease, demographic characteristics and disease-related variables (including disease stage, chemotherapy, site of lymph involvement, etc.) were extracted from the records of 387 patients with Hodgkin's lymphoma. To investigate the prognostic factors that affect the recurrence of disease the Cox model and RSF were implemented. Moreover, their performance based on the C-index, IBS, and predictor error rate of the two models were compared Data analysis was implemented by using R4.0.3 software (survival and RandomForestSRC packages).

Results: The results of the Cox model showed that LDH (P=0.001) and classical lymphoma classification (P<0.001) were associated with an increased risk of relapse in patients. However, the results of the RSF model showed that the important variables affecting the recurrence of disease were the stage of disease, chemotherapy, classical lymphoma classification, and hemoglobin, respectively. Also, the RSF model showed a higher (c-index=84.9) than the Cox model (c-index=57.6). Furthermore, the RSF model revealed a lower error rate predictor (0.09) and IBS index (0.175) than the Cox model. So, RSF has performed better than the Cox model in determining prognostic factors based on the suitability indicators of the model. Conclusion: The RSF has high accuracy than the Cox model when there is a high number of predictors and there is collinearity. It can also identify the important variables that affect the patient's survival.