Author: Yasaman Anbari, Srinivas Cheenu Kappadath, Benjamin P. Lopez, Armeen Mahvash, Ali Yousefi 👨🔬
Affiliation: University of Houston, UT MD Anderson Cancer Center 🌍
Purpose: Patient-demographics-model-based liver volumetry is well-established for determining the future liver remnant following hepatectomy. We used gold-standard CT liver segmentation to validate the performance of 10 published normal liver (NL) volumetry models in our patient population and examined their applicability to hepatocellular carcinoma (HCC) patients.
Methods: We collected gender, age, weight, height, and diagnostic CT scans of 40 patients with NL and 40 with HCC. Ground-truth CT liver volumes were determined with Contour ProtégéAI+ (MIM Software). Model-based liver volumes were compared to CT-based volumes using Bland-Altman analysis (mean bias, mean absolute-bias, 95% limits-of-agreement [LOA]). The percentage of cases with clinically acceptable performance (mean absolute-bias <10%) was calculated.
Results: The performance of 10 models for NL volumetry ranged -20% to -9% mean bias, 18-24% mean absolute-bias, and ±45-54% 95%-LOA, with 23-43% of cases showing mean absolute-bias <10%. Their performance in HCC patients ranged -23% to 5% mean bias, 17-27% mean absolute-bias, and ±39-51% 95%-LOA, with 13-53% of cases showing mean absolute-bias <10%. Vauthey-BSA model had the highest agreement with CT-based volumes, with <10% bias in 43% NL and 53% HCC patients, but still had a mean absolute-bias of 19% and 17% in NL and HCC patients, respectively. All models over-estimated liver volumes <1500 mL and under-estimated liver volumes >2000 mL in both NL and HCC patients. Vauthey-BSA showed a mean bias of +23% for NL volumes <1500 mL, +5% for 1500-1900 mL, and -16% for >1900 mL.
Conclusion: Model-based liver volumetry had comparable performance in both NL and HCC patient populations. However, they were generally unreliable for liver volumes <1500 mL and >1900 mL, with most cases (>50%) exceeding greater than ±10% error. Ongoing work with a larger patient population includes the development of machine-learning algorithms to improve liver volumetry and the correlation of clinical complications with model performance.