Objective:

To evaluate the diagnostic performance of a machine learning model based on radiomics features extracted from T2-weighted imaging fat suppression (T2WI-FS) images in preoperatively predicting metastasis of axillary lymph nodes (ALN)in breast cancer patients.

Methods:

In this retrospective study

68 pathologically confirmed breast cancer patients with 171 ALNs (101 metastatic ALN and 70 non-metastatic ALN) were enrolled

the metastatic status of which were confirmed by histopathology. Based on the manually segmented three-dimensional volumes of interest (VOI) on the T2WI-FS images of selected lymph nodes

107 radiomics features

including first-order statistics

shape- and size-based features

texture features were extracted. Using hold- out cross-validation scheme (8∶2)

F-test based select K best and the least absolute shrinkage and selection operator (LASSO) algorithms were applied to reduce features redundancy. The K-nearest neighbor (KNN)

support vector machine (SVM) and logistic regression (LR) classifiers were respectively implemented to build the prediction model. The performance was evaluated by receive operative characteristic (ROC) curves analysis with sensitivity

specificity

and precision

recall

F1-score calculated for each model

and the DeLong test was used to compare the diagnostic efficacy of different prediction models.

Results:

For each VOI

a total of 107 radiomics features were extracted. Among the 107 extracted quantitative radiomics features

6 most informative features were eventually selected for model construction

including one morphological feature (surface volume ratio) and five textural features (dependence entropy

run entropy

dependence non uniformity normalized

run percentage

large area high gray level emphasis). AUC values of LR

KNN and SVM models were 0.88

0.86 and 0.86

respectively. DeLong test showed that all differences were not statistically significant (

P

＞0.05). The efficiency o

f LR model might be slightly higher. The sensitivity

specificity

accuracy

recall rate and F1 values of the LR model in the test set were 0.86

0.86

0.80

0.86 and 0.83

respectively.

Conclusions:

Radiomics features extracted from lymph nodes based on T2WI-FS images can provide extra valuable information to preoperatively evaluate ALN metastatic status in patients with breast cancer.