Assessing microvascular invasion in HBV-related hepatocellular carcinoma: an online interactive nomogram integrating inflammatory markers, radiomics, and convolutional neural networks

Yun Zhong ^1,2,3,4, Lingfeng Chen ^1,2,3,4, Fadian Ding ^1,2,3,4

⁰Department of Hepatobiliary and Pancreatic Surgery, The First Affiliated Hospital, Fujian Medical University, Fuzhou, China.

Frontiers in Oncology +

|

October 1, 2024

View abstract on PubMed

Summary

A new nomogram accurately predicts microvascular invasion (MVI) in hepatitis B virus-related hepatocellular carcinoma (HBV-HCC) patients before surgery. This tool combines clinical data, inflammation markers like neutrophil-to-lymphocyte ratio (NLR), and MRI radiomics for improved prediction, aiding in better patient management.

Area Of Science

Hepatocellular Carcinoma Research
Medical Imaging and Diagnostics
Oncology

Background

Early recurrence of hepatocellular carcinoma (HCC) significantly reduces patient survival.
Microvascular invasion (MVI) is a critical factor affecting survival and metastasis in HBV-related HCC.
Accurate preoperative prediction of MVI is essential for effective treatment planning.

Purpose Of The Study

To develop and validate a web-based nomogram for the preoperative prediction of MVI in HBV-HCC patients.
To integrate clinical factors, inflammation markers, and MRI-derived radiomics and deep learning features into a predictive model.
To enhance the accuracy of MVI risk stratification in HBV-HCC.

Main Methods

A cohort of 173 HBV-HCC patients was divided into training (70%) and validation (30%) sets.
Pyradiomics and 3D ResNet were used to extract MRI signatures.
LASSO regression selected key features, including serum AFP, AST, NLR, radiomics, and deep features, for the nomogram.

Main Results

The neutrophil-to-lymphocyte ratio (NLR) showed a positive correlation with MRI radiomics and deep learning features.
A combined model incorporating NLR and MRI features demonstrated superior predictive performance compared to clinical or radiomics-only models.
The combined model achieved high accuracy with AUCs of 0.911 (training) and 0.907 (validation), and C-indices of 0.926 (training) and 0.917 (validation).

Conclusions

The developed nomogram effectively predicts individualized MVI risk in HBV-HCC patients preoperatively.
Integration of NLR and MRI-based features significantly improves MVI prediction accuracy.
This tool offers a valuable non-invasive method for assessing MVI risk, guiding surgical decisions and improving patient outcomes.

Keywords:

convolutional neural network hepatocellular carcinoma inflammation marker microvascular invasion radiomics