Predictive Modeling and Biomarkers Using Machine Learning and Radiomics
Predictive models at the tissue level.
Using imaging biomarkers to predict future progression of cancer, as well as response to treatment, is of central importance in cancer imaging and the main focus of the rapidly growing field of Radiomics. Our group has been engaged in use of machine learning methods for predicting cancer infiltration, patient survival and response to treatment for glioblastoma, which is the most aggressive brain cancer. In particular, we have developed predictive models of peri-tumoral infiltration in edematous regions, thereby highlighting peri-tumoral heterogeneity and its relationship to tumor progression. Fig. 1 shows a representative example of a heat map, with red regions being more likely to present future recurrence.
The overall goal of this line of research is to guide extensive, yet targeted peri-tumoral resection and focused radiation therapy in regions relatively more likely to recur early.
Predictive Models and Biomarkers at the Patient Level.
Machine learning techniques are also being used, in our laboratory, to construct predictors of patient outcome. In particular, we have found that baseline multi-parametric imaging signatures predict patient survival with very promising accuracy, as shown in Fig. 2.
These predictors not only promise to influence patient management, but also to allow clinical trials to substantially improve their ability to detect treatment effects by allowing us to select relatively homogeneous sets of patients into a trial, and hence increase power.
1. Akbari H, Macyszyn L, Da X, Bilello M, Wolf RL, Martinez-Lage M, Biros G, Alonso-Basanta M, O'Rourke DM, Davatzikos C: Imaging Surrogates of Infiltration Obtained Via Multiparametric Imaging Pattern Analysis Predict Subsequent Location of Recurrence of Glioblastoma. Neurosurgery 2016, 78(4):572-580.
2. Macyszyn L, Akbari H, Pisapia JM, Da X, Attiah M, Pigrish V, Bi Y, Pal S, Davuluri RV, Roccograndi L et al: Imaging patterns predict patient survival and molecular subtype in glioblastoma via machine learning techniques. Neuro-oncology 2016, 18(3):417-425.