Hepatocellular carcinoma (HCC) is a malignant tumor that is common worldwide. It is characterized by high incidence and mortality rates. Interventional therapy is a minimally invasive treatment for HCC that offers diverse methods that cover different stages. Because of the significant heterogeneity of tumors, even at the same stage, the effectiveness of interventional therapy can vary greatly, which makes it difficult for clinicians to determine the optimal treatment plan before treatment. Increasing evidence suggests that tumor-related imaging characteristics are correlated with biological functions and can be used to predict different subtypes of HCC and reflect their heterogeneity. In recent years, artificial intelligence (AI) has received widespread attention and been applied widely. AI can automatically extract features from medical images, objectively quantifying low-dimensional to high-dimensional information about tumors, which helps to directly or indirectly predict prognostic stratification and treatment response to interventional therapy. Furthermore, when AI integrates high-dimensional quantifiable information from imaging data with multimodal clinical and molecular data, its accuracy and interpretability improve significantly. Although image-derived AI models have achieved good performance and have broad prospects for application in the prognosis and treatment of HCC, their clinical implementation has limitations, including data and imaging standardization, model interpretability, and the need for multicenter validation. This review summarizes the latest advancements in medical image-driven AI in the prognostic stratification and efficacy prediction of interventional therapy for HCC, and outlines the main challenges that need to be addressed and good prospects for application.

Ultrasmall silica nanoparticles, as one type of nanocarriers featured by excellent biocompatibility and efficient renal clearance, are of rapidly growing interest for biomedical applications, particularly in oncology. Undesirably, the intrinsic issues of low site-targeting capability, short circulation time, and limited functionalities of ultrasmall silica nanoparticles severely impede their widespread application in the biomedical domain. Recent researches on surface modification for improved physical properties, enhanced site-specific abilities and multimodality imaging have been continuously emerging, which provide the prerequisite for possible application in the integration of diagnosis and treatment. On this basis, this review summarizes the most widely used synthesis approaches for well-ordered ultrasmall silica nanoparticles with uniform diameter and tunable pore size, and simultaneously highlights the diverse surface functionalization for versatile purposes and biomedical applications, including site-targeted delivery of drugs, stimuli-responsive cargo release, real-time bioimaging as well as cancer theranostics. Finally, the challenges of ultrasmall silica nanoparticles in oncology are further discussed with the aim of promoting their future clinical application.