Detection of hepatocellular carcinoma's microvascular invasion at the preoperative CT scan: Artificial intelligence meets radiomics. (March 2023)

• Record Type: Journal Article
• Authors: Famularo, S.; Donadon, M.; Penzo, C.; Bortolotto, M.; Maino, C.; Marescaux, J.; Diana, M.; Romano, F.; Giuliante, F.; Ardito, F.; Grazi, G.L.; Bernasconi, D.; Torzilli, G.
• Is Part Of: Digestive and liver disease. Volume 55(2023)Supplement 1
• Page Start: S6
• Publication Date: 2023-03
• Subjects: Digestive organs -- Diseases -- Periodicals; Liver -- Diseases -- Periodicals; 616.33005
• DOI: 10.1016/j.dld.2023.01.010
• Languages: English
• ISSNs: 1590-8658

Abstract

Abstract : Introduction: Microvascular invasion(MVI) is the main risk factor for overall mortality and recurrence after surgery for hepatocellular carcinoma(HCC). Its diagnosis can be made only postoperatively on the histological specimen. Aim: to train machine-learning models to predict MVI on preoperative CT scan (fig.1). Methods: Clinical data and 3-phases CT scans were retrospectively collected among 4 Italian centres. After an initial manual segmentation, an algorithm was developed to automatically identify the liver and the tumor on CT scans. Radiomics features were automatically extracted from the tumoral, peritumoral and healthy liver areas in each phase. Principal component analysis (PCA) was performed to reduce the dimensions of the dataset. Data were divided between training (70%) and test (30%) sets. Random-Forest (RF), fully connected Artificial neural network (neuralnet) and extreme gradient boosting (XGB) models were fitted to predict MVI. Hyperparameters tuning was made to reduce the out-of-bag error. Results: Between 2008 and 2022, 218 consecutive preoperative CT scans of patients affected by HCC and submitted to surgery were collected. At the histological specimen 33.02% patients had MVI. The Jaccard index between manual and algorithm segmentations was 90%. First and second order radiomics features were extracted, obtaining 672 variables per patient. PCA selected 58 dimensions explaining >95% of the variance. After standardization and normalization, RF, neuralnet and XGB were fitted to predict the presence of MVI. Tuning parameters were: 1) RF: n.tree=500, mtry=30; 2) Neuralnet: 2 hidden layer with 40 and 20 neurons, learning rate= 0.001, threshold for termination= 1%, activation function= sigmoid; 3) XGB: nrounds = 100, max_depth = 3, eta = 0.3. The models were then fitted in the testset to estimate prediction accuracy by confusion-matrix. RF was the best performer (Acc=98.4%, 95%CI: 0.91-0.99, Sens: 95.2%, Spec: 100%, PPV: 100% and NPV: 97.7%, fig.2). Conclusion: RF model predicted automatically MVI with a never-before reached accuracy