Click here to


Are you sure ?

Yes, do it No, cancel

Meaningful Segregation of Left-Breast Patients Suitable for Breath-Hold Radiotherapy Using a Machine Learning Method

L Yuan*, M Rosu , Virginia Commonwealth University Medical Center, Richmond, VA


(Sunday, 7/29/2018) 4:00 PM - 5:00 PM

Room: Exhibit Hall | Forum 3

Purpose: (1) Develop a mathematical model to estimate heart and lung dose sparing achievable with breath-hold(BH) versus free-breathing(FB) in treating left-sided breast cancer; (2) Use it to segregate those cases for which the sparing is meaningful.

Methods: Treatment plans for 40 patients (20 FB; 20 BH, using ABC) who underwent left whole breast RT were analyzed (opposing tangential beams, with clinical borders). The whole breast, ipsilateral lung and heart were contoured. A mathematical model was trained from the clinical plans to predict heart and lung DVHs based on patient anatomical features extracted from the structures (geometrical relationship between the OAR and breast; volumes and shapes of heart, lung and breast). A principal component analysis method was employed; the significant factors were identified by a stepwise regression method. The model was validated by leave-one-out cross validation. The dosimetric thresholds of interest are: ipsilateral lung: V20=35%, V30=20%, heart: mean heart dose MHD=4 Gy, V50>10 cc. When applied prospectively, first the estimated model verifies how the dosimetric quantities of interest compare with the threshold values. The following criterion in the decision chain is whether the benefit from DIBH is at least 5% for volume and 1 Gy for dose for the organ of interest.

Results: The significant anatomical metrics associated with both ipsilateral lung and heart dose are the left-right distance between lung and breast and the relative spanning angle of the OARs on lung; combined determination coefficients for the lung: 0.77, heart: 0.66. The correlation between the predicted and the actual plan dosimetric indices are: mean lung dose MLD - 0.74, and MHD – 0.46.

Conclusion: Significant anatomical metrics associated with heart and lung dose in whole breast RT are identified using a systematic machine learning method. The method is applied prospectively to identify patients that can benefit from the BH technique.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email