查看文章

Results

Nine of ten of the NR had deceased by the end of the study and all ten patients classified as R were still alive. The RF model selected GLCM energy and GLSZM gray-level variance as the most prominent radiomic features for response classification, while aLASSO selected only GLCM energy. Both models achieved AUC of 0.81 with a 95% confidence interval of [0.594–1]. AUC of the clinical data-only model was 0.65.

Conclusion

This study is the first to utilize daily low field MR images to predict pancreas tumor response to radiation. We have established that it is not only feasible to apply radiomic analysis to these images, but that there appears to be a signal related to treatment response that is seen in two independent supervised machine learning methods. The knowledge of disease response before follow-up may grant physicians precious time to modify the care path when possible. Further research is …