Combining Biology-based and MRI Data-driven Modeling to Predict Response to Neoadjuvant Chemotherapy in Patients with
Casey E Stowers1, Chengyue Wu1, Zhan Xu1
1From the Oden Institute for Computational Engineering and Sciences, The University of Texas at Austin, Austin, Tex (C.E.S., C.W., J.I.T., T.E.Y.); Chandra Family Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, Tex (S.K., J.I.T.); Livestrong Cancer Institutes, The University of Texas at Austin, Austin, Tex (T.E.Y.); Departments of Imaging Physics (C.W., Z.X., J.B.S., J.M., T.E.Y.), Abdominal Imaging (G.M.R.), Breast Imaging (C.W., G.M.R.), Breast Medical Oncology (C.Y.), Biostatistics (C.W.), and Institute for Data Science in Oncology (C.W.), The University of Texas MD Anderson Cancer Center, Houston, Tex; and Departments of Biomedical Engineering (C.W., T.E.Y.), Diagnostic Medicine (J.I.T., T.E.Y.), and Oncology (T.E.Y.), The University of Texas at Austin, 107 W Dean Keeton St, Stop C0800, Austin, TX 78712.
This study combined deep learning and a biology-based model to predict triple-negative breast cancer response to neoadjuvant chemotherapy (NAC) using MRI data. The integrated model accurately predicted tumor changes, aiding treatment decisions before therapy begins.
Area of Science:
- Oncology
- Biomedical Engineering
- Radiology
Background:
- Locally advanced triple-negative breast cancer (TNBC) requires effective neoadjuvant chemotherapy (NAC).
- Predicting patient response to NAC before treatment is crucial for personalized therapy.
- Current prediction methods may not fully capture tumor dynamics during NAC.
Purpose of the Study:
- To develop and validate a deep learning and biology-based mathematical model for predicting TNBC response to NAC.
- To utilize pre-treatment MRI data to estimate tumor evolution during NAC.
- To assess the model's accuracy in predicting total tumor volume (TTV) and total tumor cellularity (TTC).
Main Methods:
- A retrospective study using data from the ARTEMIS trial (NCT02276443).
- Construction and patient-specific calibration of a biology-based mathematical model of tumor response.
- Application of a convolutional neural network (CNN) to link calibrated model parameters with pre-treatment MRI data.
- Evaluation of CNN performance in predicting TTV, TTC, and tumor status.
Main Results:
- The integrated model demonstrated high concordance with measured changes in TTC (CCC=0.95) and TTV (CCC=0.94).
- CNN predictions showed an AUC of 0.72 for predicting tumor status at surgery.
- The model effectively used pre-NAC MRI data to predict spatial and temporal tumor evolution.
Conclusions:
- Deep learning integrated with a biology-based model shows significant promise for predicting TNBC response to NAC.
- This approach enables personalized treatment by providing pre-treatment response predictions.
- The model offers a non-invasive method to forecast tumor dynamics using only initial MRI scans.
