Radiomics analysis of ultrasonic image predicts sensitive effects of microwave ablation in treatment of patient with benign breast tumors. (July 2022)

• Record Type: Journal Article
• Authors: Li, Guanghui; An, Chao; Yu, Jie; Huang, Qinghua
• Is Part Of: Biomedical signal processing and control. Volume 76(2022)
• Publication Date: 2022-07
• Subjects: breast benign tumor -- microwave ablation -- predictive model -- volume reduction ratio -- radiomics signatures -- retrospective study; BBT breast benign tumor -- MWA microwave ablation -- VRR volume reduction ratio; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103722
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An ultrasound radiomics method combined with clinical characteristics-based model predicts therapeutic effect. A retrospective study predicts volume reduction ratio on benign breast tumors following microwave ablation. Clinical characteristics and ultrasound images were collected from 160 consecutive patients who underwent microwave ablation. Support vector machine, XGBoost and multi-layer perceptron are tested as classifies to output the prediction results. Totally 846 radiomics signatures are utilized to construct the models of prediction. Abstract: Breast tumors are the most commonly encountered complaint among adult females. Predicting the volume reduction ratio (VRR) following microwave ablation (MWA) in patients with benign breast tumors (BBT) is critical for assessing therapeutic success. Radiomics features derived from medical imaging have a strong prognostic predictive value. The purpose of this work is to develop and evaluate an ultrasound radiomics combined with clinical characteristics (URCC)-based model that can predict VRR following MWA. This retrospective study was authorized by an institutional review board, and each recruited patient signed a written informed consent form for the operation. Clinical characteristics and ultrasound images were collected from 160 consecutive patients with BBT who underwent MWA. Radiomics signatures is extracted utilizing the pyradiomics package. A prognostic nomogram is designed and confirmed by calibration, and a VRR prediction model is built utilizing integrated clinicopathological variables and ultrasound radiomics signatures. Positive samples are defined as those with a VRR of 25% after three months, 50% after six months, and 75% after twelve months. Our major contribution is to utilize as many radiomics features as possible (846 dimensions in our experiments) and combine them with clinical characteristics to predict the curative effect after MWA with BBT. The AUC of the URCC-based model is averagely 0.9347 in the valid dataset and 0.9545 in the test dataset, significantly outperforming models based on clinicopathological factors (AUC = 0.6182) or ultrasound radiomics signatures (AUC = 0.7636). In our dataset, the proposed model incorporating clinicopathological variables and ultrasound radiomics outperforms models utilizing only clinicopathological variables or only ultrasound radiomics in predicting VRR following MWA of benign breast tumors.