An Investigative study of the overlap volume histogram predictive capabilities in left breast tangential field treatment

Knowledge-based planning (KBP) has shown their ability to reproduce the experiences and to transfer knowledge. However, challenges persist in data availability and adaptability across different institutions. This study explores the Overlap Volume Histogram (OVH) as a dose prediction metric for left breast cancer treatment. The OVH curves for heart and lung of 93 patients treated for left breast cancer in one institution were computed using MATLAB. Overlapping volume thresholds (R5 %, R10 %, R30 %, R50 %) were interpolated. Geometric metrics of heart/lung were correlated with dose parameters (D2, D10, D20, D50, Dmean). Concerning the lung, additional volumetric parameters (V5, V20) were analysed. Pearson's correlation and various regression models were applied. The dataset was then reduced in 10-patient increments to find the minimum reliable sample size, then split into 70 patients for training and 20 for test. RMSE, slope, and intercept of linear models were computed. The Pearson analysis revealed the strongest correlation for the heart between R10_Dmean (r = −0.574, p < 0.001). For the lung, strong correlations were found between R30_Dmean (r = −0.714, p < 0.001) and R30_V20 (r = −0.686, p < 0.001). Regression analysis showed for both organs, that the linear model parameters were the most stable when reducing patient numbers. The heart model exhibited slope values (–0.12 to –0.17) and intercept values (6.2 to 7.6). For the lung, the R30_Dmean model showed slope values (–0.15 to –0.20) with intercepts (11.6 to 13.7), while the R30_V20 model had steeper slopes. Prediction models for the 20 patients indicated heart and lung Dmean (Gy) predictions had RMSE (0.78 and 1.03), while for the lung V20Gy ( %) prediction yielded RMSE of (2.80). The OVH shows potential as a stable prediction tool for left breast tangential treatments across patient subsets with minimum required data. Despite moderate predictive capabilities, the approach shows promise for optimizing dose reduction to organs at risk.