Towards integration of 64Cu-DOTA-trastuzumab PET-CT and MRI with mathematical modeling to predict response to neoadjuvant therapy in HER2 + breast cancer

Abstract

While targeted therapies exist for human epidermal growth factor receptor 2 positive (HER2 +) breast cancer, HER2 + patients do not always respond to therapy. We present the results of utilizing a biophysical mathematical model to predict tumor response for two HER2 + breast cancer patients treated with the same therapeutic regimen but who achieved different treatment outcomes. Quantitative data from magnetic resonance imaging (MRI) and ⁶⁴Cu-DOTA-trastuzumab positron emission tomography (PET) are used to estimate tumor density, perfusion, and distribution of HER2-targeted antibodies for each individual patient. MRI and PET data are collected prior to therapy, and follow-up MRI scans are acquired at a midpoint in therapy. Given these data types, we align the data sets to a common image space to enable model calibration. Once the model is parameterized with these data, we forecast treatment response with and without HER2-targeted therapy. By incorporating targeted therapy into the model, the resulting predictions are able to distinguish between the two different patient responses, increasing the difference in tumor volume change between the two patients by > 40%. This work provides a proof-of-concept strategy for processing and integrating PET and MRI modalities into a predictive, clinical-mathematical framework to provide patient-specific predictions of HER2 + treatment response.

Figure 1

Schematic of the integrated mathematical-experimental approach employed in the study. Prior to NAT, several imaging and biopsy data types were collected for each individual patient, including ¹⁸F-FDG PET-CT, but it is not being utilized in this current study. At a midpoint of NAT, follow-up MRI scans were also performed. At surgery, tissue was collected and sent to pathology for evaluation. The first two sets of data (pre-NAT and at the midpoint of NAT) were used to calibrate the mathematical model’s parameters for each individual patient. With the patient-specific parameters, the model was reinitialized at each patient’s second imaging session data and run forward to predict tumor status at the time of surgery. Then the model’s predictions were compared to the clinical outcome determined at the time of surgery.

Figure 2

Example MRI and PET data for a central slice from both patients with the tumor region boxed in red. Note that MRI and PET data were acquired in the prone position as opposed to the usual supine position using our novel breast support device (see text for details), and these example images are prior to intra-scan registration that aligns the data to one common space for modeling. Note that in the CE-MRI data (a,e), the tumor enhances more than the surrounding tissues. The DW-MRI data is represented here by the calculated apparent diffusion coefficient map (overlaid on an anatomical image) in (b,f). Note that low apparent diffusion coefficients (blue) are indictive of areas of higher cellularity. (c,g) Shows the ⁶⁴Cu-DT-PET data, while (d,h) present the ¹⁸F-FDG-PET data. Note that due to the tumor location and arm positioning, for patient 2, part of her arm is out of view in the cropped image. For both a high signal intensity was observed within the tumors for the ⁶⁴Cu-DT-PET data, whereas for the ¹⁸F-FDG-PET data the signal was not as strong for patient 1 for this slice. (While the ¹⁸F-FDG-PET data was collected with the data in this study and is part of the image processing pipeline, it was not used for this modeling study).

Figure 3

Example images of data processing for intra-scan registration of PET data to MRI data (upper panel) and generation of tumor ROIs (lower panel). Upper panel, deformable intra-scan registration for PET data to MRI data. Note that these images were all acquired in the prone position as opposed to the usual supine position using our novel breast support device (see text for details). The top row depicts a central slice for patient 1 at baseline for the CE-MRI (a), ⁶⁴Cu-DT-PET (b), and the ⁶⁴Cu-DT PET-CT (c). After the CT data is trimmed to only include breast tissue (i.e., the chest cavity and breast support was removed from the original images), the CT images were registered to the CE-MRI data using a fully deformable registration algorithm. (d–i) depict the resulting overlap between the CE-MRI data and registered CT-PET data for three central slices (green represents the CE-MRI data and pink represents the cropped ⁶⁴Cu-DT PET-CT and ⁶⁴Cu-DT-PET for (d–f) and (g–i), respectively). Lower panel, example images for generating tumor ROIs from CE-MRI data using the FCM method (patient 2). (j) Depicts a central slice of the CE-MRI data. (k) Shows a manually drawn, conservative ROI (red) on the CE-MRI data. (l) Depicts the FCM generated ROI (red).

Figure 4

The central slide of each tumor for both patients are shown to depict the tissue segmentation (a,e), number of tumor cells (b,f), and the normalized estimates of the systemic (c,g) and targeted (d,h) therapies used in the model. The tissue segmentation identifies the tumor (dark area), fibroglandular (white), and adipose (grey) tissues within the breast. For the remaining columns, the parameter maps (i.e., the colored pixels) are overlaid on anatomical images of the breasts (grey). At surgery, patient 1 was designated as a pCR, while patient 2 had residual disease and was designated as non-pCR. Both patient’s tumor exhibit areas of high cellularity, and the approximate drug distribution of the systemic therapies have similar intensities. Comparing the targeted distribution maps, patient 1 appears to have had greater drug distribution compared to patient 2 for this central slice.

Figure 5

Central slice examples of the predicted tumor cellularity by the MRI-based model compared to the PET/MRI-based model. The number of tumor cells is overlaid (color) on an anatomical image of the breast (grey). At surgery patient 1′s (left column) tumor response was designated as pCR and for patient 2 (right column) the tumor response was designated as non-pCR. Notice that the PET/MRI-based model predicts overall smaller tumors and for patient 1 specifically, lower overall cellularity.

Figure 6

Tumor response curves for total cellularity predicted by the MRI-based (blue curve) and PET/MRI-based model (orange curve) for each patient. Each panel depicts the predictions for total tumor cellularity from the time of scan 2 (day 0) to the time of surgery (end of simulation) for patient 1 (panel a) and patient 2 (panel b). Note that patient 1 received two doses of therapy between her second scan and surgery, while patient 2 received three doses of therapy during that time. The period of tumor regrowth after the completion of NAT occurs during the time patients were no longer receiving systemic therapy. Comparing the timing of each patient’s treatment schedule, patients 1 and 2 underwent surgery 73 days and 92 days, respectively, after their second scan. From their last cycle of therapy, patients 1 and 2 underwent surgery after 38 and 36 days, respectively. While both models predicted an oscillatory behavior in relation to when therapy was delivered, the MRI-based model (blue curves) predicted greater tumor regrowth during the refractory periods than the PET/MRI based model (orange curves). We conjecture this reflects the modified proliferation due to targeted therapy incorporated into the model.