Towards survival prediction of cancer patients using medical images

Survival prediction of a patient is a critical task in clinical medicine for physicians and patients to make an informed decision. Several survival and risk scoring methods have been developed to estimate the survival score of patients using clinical information. For instance, the Global Registry of Acute Coronary Events (GRACE) and Thrombolysis in Myocardial Infarction (TIMI) risk scores are developed for the survival prediction of heart patients. Recently, state-of-the-art medical imaging and analysis techniques have paved the way for survival prediction of cancer patients by understanding key features extracted from Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) scanned images with the help of image processing and machine learning techniques. However, survival prediction is a challenging task due to the complexity in benchmarking of image features, feature selection methods, and machine learning models. In this article, we evaluate the performance of 156 visual features from radiomic and hand-crafted feature classes, six feature selection methods, and 10 machine learning models to benchmark their performance. In addition, MRI scanned Brain Tumor Segmentation (BraTS) and CT scanned non-small cell lung cancer (NSCLC) datasets are used to train classification and regression models. Our results highlight that logistic regression outperforms for the classification with 66 and 54% accuracy for BraTS and NSCLC datasets, respectively. Moreover, our analysis of best-performing features shows that age is a common and significant feature for survival prediction. Also, gray level and shape-based features play a vital role in regression. We believe that the study can be helpful for oncologists, radiologists, and medical imaging researchers to understand and automate the procedure of decision-making and prognosis of cancer patients.

Automated summary

Survival prediction is a critical task in clinical medicine, and this article evaluates the performance of different features and methods for predicting the survival of cancer patients. The study highlights the importance of age as a significant feature for survival prediction.