Artificial Intelligence Model Developed to Better Predict the Early Recurrence of Hepatocellular Carcinoma - Oncology Nurse Advisor

Researchers developed an artificial intelligence model system that was effective at helping healthcare professionals identify patients with hepatocellular carcinoma (HCC) who were most at risk of early recurrence, potentially improving surveillance and treatment strategies, according to a study published in The Oncologist. 

HCC is one of the leading causes of cancer deaths globally. Early intervention is vital to improve prognosis and extend survival. Treatment options include surgical resection, thermal ablation, and liver transplantation. Thermal ablation in particular has emerged as a cost-effective option with an acceptable safety profile. 

Kong and Li conducted a study in which they developed a deep learning model (a mode of artificial intelligence) to better predict early recurrence in patients with HCC based on longitudinal magnetic resonance imaging (MRI). 

Between April 2014 and May 2017, Kong and Li retrospectively recruited 289 patients with early-stage HCC who underwent thermal ablation therapy at a university hospital in China. Early recurrence was defined as lesions seen on contrast-enhanced MRI that were away from or abutting the ablated area and presenting as abnormal patterns of peripheral enhancement at the arterial phase and the washout at the delayed phase. The follow-up period lasted until September 2023. 

Among the 289 patients, 254 were used as the training cohort, of whom 20% were randomly selected to form the tuning cohort. Thirty-five patients formed the external testing cohort. 

The resulting deep learning models (i.e., the Pre and PrePost models) were compared with the Clinical model. The performance of these models as demonstrated by the area under the receiver operating characteristic curve showed that the deep learning models held significantly higher predictive power compared with the Clinical model in the external testing cohort. 

Researchers then developed the integrated DL_Clinical Model to stratify early recurrence risk among patients with HCC.The research team grouped patients into low- and high-risk categories and reported a significant difference in relapse-free survival between these two groups in the training cohort. 

“The present study found that the performance of each [deep learning] model (the Pre or PrePost models) was higher than that of the Clinical model in the external testing cohort,” Kong and Li wrote. “This suggests that subtle information which cannot be captured by the naked eye in images may be more important than clinical variables for [early recurrence] prediction.”