Background: Macrotrabecular hepatocellular carcinoma (MTHCC) has a poor prognosis and is difficult to diagnose preoperatively. The purpose is to build and validate MRI-based models to predict the MTHCC subtype.
Methods: Two hundred eight patients with confirmed HCC were enrolled. Three models (model 1: clinicoradiologic model; model 2: fusion radiomics signature; model 3: combined model 1 and model 2) were built based on their clinical data and MR images to predict MTHCC in training and validation cohorts. The performance of the models was assessed using the area under the curve (AUC). The clinical utility of the models was estimated by decision curve analysis (DCA). A nomogram was constructed, and its calibration was evaluated.
Results: Model 1 is easier to build than models 2 and 3, with a good AUC of 0.773 (95% CI 0.696-0.838) and 0.801 (95% CI 0.681-0.891) in predicting MTHCC in training and validation cohorts, respectively. It performed slightly superior to model 2 in both training (AUC 0.747; 95% CI 0.689-0.806; p = 0.548) and validation (AUC 0.718; 95% CI 0.618-0.810; p = 0.089) cohorts and was similar to model 3 in the validation (AUC 0.866; 95% CI 0.801-0.928; p = 0.321) but inferior in the training (AUC 0.889; 95% CI 0.851-0.926; p = 0.001) cohorts. The DCA of model 1 had a higher net benefit than the treat-all and treat-none strategy at a threshold probability of 10%. The calibration curves of model 1 closely aligned with the true MTHCC rates in the training (p = 0.355) and validation sets (p = 0.364).
Conclusion: The clinicoradiologic model has a good performance in diagnosing MTHCC, and it is simpler and easier to implement, making it a valuable tool for pretherapeutic decision-making in patients.
Keywords: Decision-making; Macrotrabecular hepatocellular carcinoma; Magnetic resonance imaging; Nomograms.
© 2022. The Author(s).