Description

1. Research Background Primary liver cancer, predominantly comprising hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC), ranks as the third leading cause of cancer-related mortality. In China, approximately 70% of liver cancer patients are diagnosed at intermediate or advanced stages, where localized therapies such as surgery and ablation yield limited efficacy, and targeted therapies exhibit low overall response rates in advanced cases. This may be attributed to the heterogeneity of liver cancer, which manifests at multiple molecular levels-including genomic, transcriptomic, and metabolomic variations-both across individuals and within individual tumors. Such heterogeneity leads to divergent therapeutic responses and clinical outcomes among patients with identical pathological types and stages. Therefore, there is an urgent need to develop preoperative diagnostic methods capable of early assessment and prediction of tumor heterogeneity to guide precision clinical decision-making.

Ultrasound is the preferred imaging modality for liver cancer screening due to its cost-effectiveness, safety and widespread clinical adoption. Contrast enhanced ultrasound (CEUS), the secondary guideline-recommended imaging technique for liver cancer diagnosis, offers economic and low-risk advantages compared to first-line recommendations like dynamic contrast-enhanced CT or MRI. Liver elastography has become a standard clinical tool for assessing cirrhosis. Photoacoustic imaging (PAI), an emerging non-invasive functional imaging technology, enables visualization of specific molecules based on their spectroscopic characteristics at designated wavelengths. Extensive studies have demonstrated the significant value of combined photoacoustic/ultrasound imaging in the diagnosis and prognostic evaluation of various cancers, including breast cancer and melanoma.

This study aims to: (1) Conduct an observational investigation combining CEUS, elastography, and superb microvascular imaging (SMI) to collect imaging data; (2) Collect tumor specimens from participants for investigating heterogeneous protein characteristics in primary liver cancer organoids using PAI; (3) Analyze peripheral venous blood samples to study transcriptomic profiles. Artificial intelligence (AI) will be employed to establish prognostic models integrating ultrasound radiomics and tumor heterogeneity multi-omics features, providing novel insights for precision diagnosis and treatment of liver cancer.

2. Research workflow

1. Construction of a diagnostic and prognostic model for liver cancer based on multimodal imaging Ultrasound examinations will be performed by sonographers with over five years of abdominal ultrasound experience. All the imaging data will be recorded.During the procedure, patients will assume a supine position with a left lateral decubitus position. Conventional ultrasound, SMI, shear wave /strain elastography, and CEUS will be conducted. Baseline data will be collected preoperatively, followed by postoperative/post-conversion therapy follow-ups at 1/2/3/4/5/6, 9/12/15/18/21/24, and 30/36/42/48/54/60 months to gather additional imaging and clinical data.
2. Screening of prognosis-related heterogeneous multi-omics features and development of risk assessment models in HCC/ICC Peripheral venous blood (6-10 mL) will be collected from HCC/ICC patients. Whole blood and serum samples will be preserved, with 3 mL of whole blood and serum stored at -80°C. Another 3 mL of whole blood will be lysed in 9 mL TRIzol and stored at -80°C for multi-omics analysis. Tumor specimens will also be preserved for multi-omics studies and organoid-based heterogeneous protein characterization. Prognosis-associated transcriptomic and proteomic features will be screened based on patient outcomes, and prognostic risk models will be constructed by integrating these multi-omics profiles with clinical characteristics.
3. Development of a predictive model integrating multimodal imaging and tumor heterogeneity multi-omics features via novel knowledge transfer and model training strategies First, transfer learning method will be applied to adapt basic perceptual capabilities from HCC/ICC multimodal imaging data. Furthermore, deep interactive fusion of radiomic features and tumor heterogeneity multi-omics data will be performed to establish predictive models for degree of HCC/ICC differentiation, MVI and prognosis.