Artificial intelligence can extract important features for diagnosing axillary lymph node metastasis in early breast cancer using contrast-enhanced ultrasonography

From the article

Machine learning (ML) models are vital for providing optimal solutions and revealing complex interrelationships between virables[^(23)](https://www.nature.com/articles/s41598-025-90099-9#ref-CR23). Previous studies have stated that tree-based models outperform deep learning (DL) models on tabular data[^(24)](https://www.nature.com/articles/s41598-025-90099-9#ref-CR24), because imaging features are regarded as tabular data. A light-gradient boosting machine (LightGBM) with feature selection, a representative tree-based model, has been widely applied for evaluating tabular datasets and extracting feature importance for each parameter[^(25)](https://www.nature.com/articles/s41598-025-90099-9#ref-CR25)^(,)[^(26)](https://www.nature.com/articles/s41598-025-90099-9#ref-CR26)^(,)[^(27)](https://www.nature.com/articles/s41598-025-90099-9#ref-CR27)^(,)[^(28)](https://www.nature.com/articles/s41598-025-90099-9#ref-CR28). Attempts have also been made to predict ALN metastasis in breast cancer based on conventional US images of primary breast cancer tumour[^(29)](https://www.nature.com/articles/s41598-025-90099-9#ref-CR29). However, to date, no reports have analysed CEUS using these ML models have been published.

Based on this hypothesis, important features can be extracted in CEUS using ML models. Therefore, we developed a bimodal model to predict ALN metastasis in patients with early breast cancer by integrating CEUS images with annotated imaging features. This study aimed to evaluate the performance of this model and to extract crucial features for predicting ALN metastasis.