A CNN-based predictive model using grayscale image-converted data enhanced breast cancer prognosis prediction by capturing complex spatial patterns in 17,438 patients.
A novel deep learning approach converting 1D clinical and environmental data into grayscale images was proposed to predict disease-free survival in breast cancer patients.
Abstract Background: In survival analysis tasks, DeepSurv, DeepHit, etc., are deep learning-based models designed to handle one-dimensional time series data. Their primary objective is to predict an individual's survival probability and its relationship with time. However, traditional one-dimensional data processing methods may overlook certain spatial structural information inherent in the data—a limitation that is particularly significant for specific types of biomedical or clinical data. Therefore, we propose an innovative approach: converting one-dimensional data into grayscale images. This conversion introduces additional spatial features, enabling the model to better capture complex patterns in the data. Our aim is to develop a predictive analysis model for the impact of atmospheric environment on breast cancer prognosis. Methods: This study is a multi-center retrospective cohort study on breast cancer, involving 17,438 breast cancer patients. The endpoint outcome is disease-free survival (DFS). The features include age, gender, HDI, history of malignant tumors, family history of breast cancer, ki67, pathological type, molecular subtype, TNM staging, O3, PM2.5, monthly average precipitation, monthly average wind speed, and monthly average temperature. Firstly, we expand the dimensionality of one-dimensional survival data by mapping it to grayscale images, making it compatible with the input format of convolutional neural networks (CNNs) for processing image data. Through this conversion, we can effectively extract spatial features using convolutional operations, thereby enhancing the model's predictive capability. Subsequently, we develop a CNN-based deep neural network model that incorporates convolutional layers to process these grayscale image data while retaining the core concepts of survival analysis. This model aims to predict an individual's relative risk value by learning deep patterns in the data and can handle prognostic data with high complexity and non-linear relationships. Citation Format: Y. Yuan, A. Yuan, W. Wang, Z. Hu, C. Zheng, Z. Zheng, W. Liang, Y. Zhang, Y. Pan, C. Zhang. A Retrospective Study on a Predictive Model to Probe the Impact of Atmospheric Environment on Breast Cancer Prognosis: A Cohort of 17,438 Patients abstract. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS3-06-15.
Yuan et al. (2026) studied this question. A CNN-based predictive model using grayscale image-converted data enhanced breast cancer prognosis prediction by capturing complex spatial patterns in 17,438 patients.