I can share your concern with the AlignRT version we have used couple years. Just a minute ago we had problematic DIBH patient for the first time in treatment, starting of the treatment thus. Entire chest wall was covered with ROI (because system asked us to increase the ROI) and still there was that error as you mentioned. This happened when we tried to acquire kV- tangential images (side pods were blocked). We had to remove kV- tangential image and add MV tangential image on the fly.
We have used ROI below and investigated its reproducibility with several IGRT studies. Based on the results, we have not seen any reasons to use breastROI as a primary setup-ROI. With that ROI below, sternum vertical reproducibility in the setup is very accurate, indicating for its part good accuracy in the BHL. We have been able to acquire daily tangential kV-images, when gantry and imagers are both blocking side cameras and typically this has no affect on image accuracy, if the deltas act as they should, without blinking. Rarely there is issues that you described, and if there is, it happens when the gantry is somewhere around 30°. Then the ROI needs to be adjusted, and if so, new orthogonal images and possibly new reference surfaces should be acquired, since we are now monitoring slightly different area.
With the larger sized breast and with the VMAT- treatments, we do a breastROI verification at the FB-setup process with the AlignRT, to see that also the soft tissue is inside thresholds after T-ROI setup. We have a data whether T -or BreastROI leads to better monitoring reproducibility with different gantry angles, but we have not investigated that. Anyhow, setup accuracy comparison with the differently shaped ROIs based on offline image+AlignRT data evaluation is planned to be published this year. If you use ROI below and spread it laterally in a problematic cases, based on the results I can say that you are not loosing the accuracy concerning BHL in the LAT image in comparison to breastROI, which was your worry. As mentioned, you may anyway need new setup surfaces based on setup images if this corrective action is done only during the treatment course. 10° tilt in the breast board gives a better conditions for the cameras to monitor the patient chest wall.
It is much possible that these are ROI- related issues, but slight but in practice significant improvements also I have been waiting from VisionRT to optimize the camera system to better monitor different kind of ROIs with all the gantry angles.