Effect of Noise Addition and Voxel Grouping Variations on Fitting Accuracy in Voxel Based Dosimetry Using [177Lu]Lu-DOTATATE Farida Nurlina, Yasmin Fauziah, Assyifa Rahman Hakim, Teuku Muhammad Zaki Yasykur Polem, Muhammad Syifa, M. Buchori Amin Kartani, Deni Hardiansyah
Medical Physics and Biophysics, Physics Department, Faculty of Mathematics and Natural Sciences, Universitas Indonesia, Depok, Indonesia
Abstract
This study investigated the effects of Poisson noise and voxel grouping size on model fitting performance in voxel based dosimetry of Hepatocellular Carcinoma (HCC) patients undergoing 177Lu DOTATATE therapy.
Analyses were performed on the liver, right kidney, left kidney, and spleen from two HCC patients. Unit voxel sizes were 7.45 cubic millimeters for patient 4 and 2.86 cubic millimeters for patient 6, with grouping dimensions ranging up to 168.11, 216.79, and 279.21 millimeters for patient 4 and 300, 97.66, and 97.66 millimeters for patient 6. Each grouping variation was fitted using monoexponential, biexponential f3b, and biexponential f3a functions. The optimal model for each voxel was selected using the lowest Bayesian Information Criterion (BIC). Fitting results were evaluated using Relative Difference (RD), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) by comparing noisy results with reference results without noise.
RMSE values varied across organs and patients after the addition of Poisson noise. In the left kidney, RMSE ranged from 10.05% to 56.42% in patient 4 and from 11.26% to 89.06% in patient 6, with minimum values observed at grouping dimensions of 66.44, 85.80, and 110.44 millimeters and 129, 42.14, and 42.14 millimeters, respectively. In the right kidney, RMSE ranged from 6.97% to 68.43% in patient 4 and from 4.63% to 119.92% in patient 6, with minimum values observed at grouping dimensions of 89.09, 115.05, and 148.09 millimeters and 126, 41.16, and 41.16 millimeters, respectively. In the liver, RMSE ranged from 12.08% to 84.22% in patient 4 and from 14.73% to 94.10% in patient 6, with minimum values observed at grouping dimensions of 167.61, 216.45, and 278.61 millimeters and 201, 65.66, and 65.66 millimeters, respectively. The spleen in patient 4 showed RMSE values ranging from 3.34% to 43.58%, with the minimum value observed at grouping dimensions of 80.03, 103.35, and 133.03 millimeters, indicating the lowest sensitivity to Poisson noise among all analyzed organs. Poisson noise and voxel grouping size influenced dose estimation accuracy, where certain grouping sizes produced lower RMSE values and smaller deviations between noisy and reference results.
Keywords: Voxel based dosimetry, HCC, Noise addition, Voxel grouping size.
