AI-driven Digital Twin in Breast Cancer Neoadjuvant Therapy: MCPS-Based Adaptive Treatment Planning and Tumor Response Prediction via Deep Learning (pp.299-315)
X. Zhou1, L. Xie1, P. Li2, J. Liu1, G. Liao1, H. Yang1,*
https://doi.org/10.30546/1683-6154.25.2.2026.2991Department of Breast Surgery, Guangxi Medical University Cancer Hospital, Nanning, Guangxi, 530201, China,
e-mail: zxmed1981@163.com, g13393646184@163.com, 202210168@sr.gxmu.edu.cn, liujianhui0131@163.com,
liaogeyi1@yeah.net
∗Corresponding author: e-mail: 930060@sr.gxmu.edu.cn
Abstract. This paper presents an AI-engineered Digital Twin module that can be incorporated into a Medical Cyber-Physical System (MCPS) and allow creating adaptable and patientcentered treatment plans. The model integrates longitudinal imaging, genomic and clinical features using a multi-modal attention-gated deep learning model in order to forecast tumor response in real-time across therapy cycles. An Model Predictive Control (MPC)-based planner takes advantage of these predictions to optimize dosage of chemotherapy and achieve balanced therapeutic efficacy and toxicity. In silico simulations on a set of virtual patients undergoing NAT are considered to evaluate the system. Findings indicate that the adaptive strategy shows greater predicted pathological complete response rates and lower final tumor volume than fixeddose regimens and does not result in more cumulative toxicity. The suggested architecture presents a scaled and risk-free route to real-time and factual optimization of cancer treatment and emphasizes the chance of shut-loop intelligent assemblies in precision oncology.
Keywords: Digital Twin, Breast Cancer, Neoadjuvant Therapy, Medical Cyber-Physical Systems (MCPS), Deep Learning, Adaptive Treatment Planning, Tumor Response Prediction, Mechanistic Learning, Control Systems.
AMS Subject Classification: 68T07, 92C50, 93C85, 68U10.