Radiomics in CT imaging for Longitudinal Analysis in Oncology: Emerging Complements, Advantages and Limitations

Abstract

Precision medicine plays a crucial role for cancer diagnosis and treatment planning. With significant advancements in the medical imaging field, new techniques are being developed to characterize diseases, predict treatment outcomes, and determine survival rates. In this context, the collection of image data over time known as longitudinal imaging has become increasingly popular in clinical oncology for its prognosis potential. Radiomics is a widely used tool to extract features from such images to analyze them. While traditional radiomic approaches such as texture or shape analysis have proven to be effective, they may lack sensitivity to certain changes, occasionally oversimplifying the complexity of the data under study. Furthermore, these methods are sometimes not stable enough to noise or image artifacts, and do not generalize well, making it challenging to compare scans over time or across different studies. In this paper, we present a review of the use of radiomics in longitudinal oncologic studies, specially focusing on its applications in CT imaging. To this end, this study aims to explore three main alternative methods to traditional radiomics: for feature extraction, topological data analysis (TDA), including persistent homology, and geodesic geometry; and for data analysis, Cox and joint statistical modeling. By capturing and analyzing more complex features, these methods offer new valuable insights into disease progression, making them strong candidates for the development of more accurate treatment planning and prognostic models in clinical oncology.