Predicting breast cancer response to neoadjuvant chemotherapy with ultrasound-based deep learning radiomics models -- dual-center study.
Authors
Affiliations (6)
Affiliations (6)
- Department of Ultrasonography, The Tenth Affiliated Hospital of Southern Medical University(Dongguan People's Hospital), Dongguan, Guangdong, 523059, China.
- Department of Ultrasonography, The Tenth Affiliated Hospital of Southern Medical University(Dongguan People's Hospital), Dongguan, Guangdong, 523059, China. [email protected].
- Department of Ultrasonography, The Affiliated Tumor Hospital of Xinjiang Medical University, Urumqi, Xinjiang, 830011, China. [email protected].
- Southern Medical University, Guangzhou, Guangdong, 510515, China.
- Guangdong Medical University, Zhanjiang, Guangdong, 524023, China.
- Department of Ultrasonography, The Affiliated Tumor Hospital of Xinjiang Medical University, Urumqi, Xinjiang, 830011, China.
Abstract
The aim of this study is to develop and validate a deep learning fusion model based on early (two-cycle) ultrasound images, using stacking fusion technology for the intratumoral and peritumoral regions, to predict early tumor response in breast cancer patients receiving neoadjuvant chemotherapy (NAC). This study enrolled breast cancer patients who received 6 to 8 cycles of neoadjuvant chemotherapy (NAC) between May 2019 and September 2023 from two medical centers. In the training set, the ResNet deep learning method was employed to construct a deep learning model based on ultrasound image data from intratumoral and peritumoral regions. A deep learning fusion model was established using stacking technology, which was tested on internal and external validation sets to calculate the areas under the receiver operating characteristic curve (AUC) and clinical decision curve (DCA). A total of 469 breast cancer patients were included in this study. The AUC values of the fusion model DLRS3 in the training set for 3 mm ROI (ROI3mm), 5 mm ROI (ROI5mm), 10 mm ROI (ROI10mm), and intratumoral region (ROIin) were 0.858, 0.872, 0.919, and 0.848, respectively. The AUC values in the internal validation set were 0.890, 0.965, 0.915, and 0.840, respectively, while in the external validation set, they were 0.844, 0.886, 0.938, and 0.905, respectively. This study successfully developed a fusion model for predicting early tumor response in breast cancer patients receiving NAC. The model combines deep learning methods based on early NAC ultrasound images and demonstrates high predictive performance.