Statistical Analysis to Determine the Predictors of Liver Segmental Hypertrophy Observed Post-Radiotherapy

Abstract

Prediction of liver segment hypertrophy based on radiotherapy (RT) dose is crucial for maximizing functional liver volume and avoiding hepatic failure after RT. We determined predictors associated with liver hypertrophy with stratification based on induction chemo (IC) and tumor location. RT planning, CT images, and 3-month-followup CTs were analyzed from 148 patients who underwent RT for primary or metastatic liver cancers. A nnUNet based model was trained (train/test = 160/40 CTs) to contour the liver segments (1, 2, 3, 4, 5-8) with accuracy assessed using Dice Similarity Coefficients (DSC). 52 features corresponding to segments 1, 2, 3, 2+3, 4, 5-8, were collected including equivalent dose to 2Gy fractions metrics-mean dose (Dmean), dose received by 95% of the volume (D95), volume spared from x gy (Vx), cancer type, tumor location, and IC status. Descriptive statistics were reported as percentage of segments showing hypertrophy under all stratification. Predictors were compared with 6 response variables using Chi-squared/Fisher-Exact test (CST/FET) and logistic regression (LR) for categorical and numerical predictors. The nnUNet model had an average DSC of 0.91 across all segments. Overall, segments 1, 4, and 5-8 showed hypertrophy in 35% of cases, and segments 2, 3 and 2+3 showed hypertrophy in 45-49% of patients. Stratification based on tumor location resulted in segment 2+3 hypertrophy in 66% of patients when the tumor was in segments 5-8. For bilobed tumors, segment 2+3 hypertrophy was observed in 34% of patients. CST/FET showed that tumor location, IC, and tumor type were significant predictors of segment 5-8 hypertrophy. Tumor location was also a significant predictor of segments 3 and 2+3 hypertrophy. In LR analysis, all segment-based dose metrics were significant predictors of segment hypertrophy except Dmean in segment 4, and D95 in segment 2 and 4. Overall, the strongest association was obtained for V35 significantly predicting for each segment hypertrophy. The mean dose for segments with hypertrophy was significantly lower (range: 15-30 Gy) than segments with atrophy (p<0.01), except for segment 4 where the mean dose was 10Gy lower but did not reach significance. IC impacts the threshold mean dose that leads to hypertrophy, with more toxic drugs reducing the mean dose threshold. Tumor location and IC significantly impact the response of segments to RT. Dose volume metrics are strong predictors of volumetric response with segment volume spared from 35 Gy being the strongest predictor.