SFMR:s företagsstipendium: Qiang Wang

Combining radiomic and body composition analysis for early readout of insufficient liver volume hypertrophy after portal vein embolization

Portal vein embolization (PVE) has been regarded as a standard procedure to induce future liver remnant (FLR) hypertrophy before extended liver resection. However, around 30% of patients who undergo PVE cannot proceed to liver resection due to insufficient FLR growth or tumor progression. This study aims to incorporate radiomics into body composition analysis to develop and externally validate a model for predicting patients who have a high risk of insufficient FLR hypertrophy after PVE.

Patients with liver cancer who underwent PVE intended for liver resection at three Swedish medical centers will be included (training cohort: ca.120 patients; two test cohorts: each ca. 70 patients). Insufficient FLR growth was defined as standardized FLR < 30% after PVE at first imaging evaluation. Radiomics features will be extracted from three body composition compartments (skeletal muscle, subcutaneous and visceral adipose) at the third lumbar vertebra (L3) level of pretreatment CT images by ‘pyradiomics’ package. Body composition segmentation will be performed using TotalSegmentator. Traditional body composition indices (e.g. skeletal muscle index) and clinicopathological factors will also be integrated for model development. L1 regularization technique (LASSO) will be used to select informative features. Classifiers of logistic regression, random forest, support vector machine, and XGBoost will be adopted to fit three models (clinical model, body composition radiomics model and combined model). The model performance will be validated in two test external cohorts using the area under the curve. The models with the highest AUC at both training and test cohorts will be selected as the ideal model.