Quantifying The Predictability Of Left Breast Surface Motion During Dibh Treatment With An External Marker

The ideal way of treating left breast Deep Inspiration Breath Hold (DIBH) is through direct breast surface matching. Many DIBH techniques, however, rely on a surrogate—external point/marker motion. We try to quantify the predictability of the breast surface motion using external point/marker. AlignRT Beam Hold system is applied to perform real-time surface matching and the external point/marker tracking simultaneously. The skin rendering of breath hold CT scan is served as the reference. During the treatment, the patient surface is monitored and registered to the reference to calculate the corresponding distance (S(t)). Radiation beam is turned on when this distance is within a preselected threshold. The external point/marker tracking is implemented by tracking the vertical amplitude of a point in the center part of the left breast skin. The real-time distance (P(t)) of the selected point to the corresponding reference point is calculated. We model S(t) as a proportioned P(t). Statistical and computational complexity analyses are conducted. 7 patients are included in this study. The coefficient for the general linear model (p value) is calculated for each patient. All p values are smaller than 0.0001 which indicates the strong correlation between S(t) and P(t)—the use of a linear model is justified. For each patient, the ratio of the linear model is calculated for the first treatment day and applied on the subsequent days. The difference between the prediction and the true S(t) is calculated. The average standard deviations of the difference over all the treatment days are 1.52 - 2.85mm for different patient, corresponding to 2.4 - 14.6% error rate for 3mm threshold. For instance, for error rate of 3.9%, the probability is 0.039 that the prediction will result beam on while the true breast surface is more than 3mm away from the reference. The biggest standard deviation is from the patient with the largest breast. This might due to the larger uncertainty of the day to day breast tissue positioning variation comparing to small breast or mastectomy patients. More data need to be collected to confirm this observation. The computational complexity of the fast ICP based surface matching algorithm is O(N), and of the point/marker tracking is O(1). Strong linear correlation has been found between left breast surface motion and external marker motion. For small breast and mastectomy patients, high prediction accuracy can be achieved using the external point/marker surrogate by using slightly larger threshold. For big breast patients, the prediction might introduce unaccepted error rate. Overall, breast surface matching is more accurate comparing to the external point/marker surrogate at the price of higher computational complexity.