Deep learning–based and BI-RADS guided radiomics model for tumour-infiltrating lymphocytes evaluation in breast cancer

Abstract

Abstract Objectives To investigate an interpretable radiomics model consistent with the clinical decision-making process and realize tumour-infiltrating lymphocytes (TILs) levels prediction in breast cancer (BC) from ultrasound images. Methods A total of 378 patients with invasive BC confirmed by pathological results were retrospectively enrolled in this study. Radiomics features were extracted guided by the breast imaging reporting and data system (BI-RADS) lexicon from the regions of interest (ROIs) segmented with deep learning models. After features were selected using the least absolute shrinkage and selection operator regression, 4 machine learning classifiers were used to establish the radiomics signature (Rad-score). Then, the integrated model was developed on the basis of the best Rad-score incorporating the independent clinical factors for TIL level prediction. Results Tumours were segmented using the deep learning models with an accuracy of 97.2%, sensitivity of 93.4%, specificity of 98.1%, and the posterior areas were also obtained. Eighteen morphology and texture-related features were extracted from the ROIs and 14 features were selected to construct the Rad-score models. Combined with independent clinical characteristics, the integrated model achieved an area under the curve of 0.889 (95% CI, 0.739-0.990) in the validation cohort, which outperformed the traditional radiomics model and achieved comparable performance with the deep learning models. Conclusions This study established a promising model for TIL levels prediction with numerable interpretable features and showed great potential to help decision-making and clinical applications. Advances in knowledge Imaging-based biomarkers have provided noninvasive ways for TIL levels evaluation in BC. Our model combining the BI-RADS-guided radiomics features and clinical data outperformed the traditional radiomics approaches and is comparable to deep learning methods.