# Discrimination of superficial lymph nodes using ultrasonography and tissue metabolomics coupled with machine learning

## TL;DR

This study combines ultrasound imaging and metabolomics with machine learning to distinguish between benign and malignant lymph nodes, offering improved diagnostic accuracy.

## Contribution

The study introduces a novel approach combining ultrasonography, metabolomics, and machine learning for lymph node classification.

## Key findings

- Ultrasonographic features like blood supply pattern and cortex elasticity help differentiate malignant from benign lymph nodes.
- Metabolomics identified distinct metabolic profiles in lymphoma and metastasis groups, with specific metabolites showing significant differences.
- Linear discriminant analysis using metabolites achieved higher recognition rates (87.4%-89.3%) compared to ultrasonography (63.1%-75.4%).

## Abstract

Diagnosing the types of malignant lymphoma could help determine the most suitable treatment, anticipate the probability of recurrence and guide long-term monitoring and follow-up care.

We evaluated the differences in benign, lymphoma and metastasis superficial lymph nodes using ultrasonography and tissue metabolomics.

Our findings indicated that three ultrasonographic features, blood supply pattern, cortical echo, and cortex elasticity, hold potential in differentiating malignant lymph nodes from benign ones, and the shape and corticomedullary boundary emerged as significant indicators for distinguishing between metastatic and lymphoma groups. Metabolomics revealed the difference in metabolic profiles among lymph nodes. We observed significant increases in many amino acids, organic acids, lipids, and nucleosides in both lymphoma and metastasis groups, compared to the benign group. Specifically, the lymphoma group exhibited higher levels of nucleotides (inosine monophosphate and adenosine diphosphate) as well as glutamic acid, and the metastasis group was characterized by higher levels of carbohydrates, acylcarnitines, glycerophospholipids, and uric acid. Linear discriminant analysis coupled with these metabolites could be used for differentiating lymph nodes, achieving recognition rates ranging from 87.4% to 89.3%, outperforming ultrasonography (63.1% to 75.4%).

Our findings could contribute to a better understanding of malignant lymph node development and provide novel targets for therapeutic interventions.