可穿戴计算机
医学
结直肠癌
计算机科学
危险分层
肿瘤科
分层(种子)
可穿戴技术
内科学
医学物理学
癌症
远程病人监护
作者
Guangyu Yao,Lei Hu,Yadong Guo,Xicheng Wang,X W Chen,Lu Liu,Jie Tang,Yi Li,Cheng Cheng,Chuan-Peng Wang,Q R Zhou,Yi Ba
标识
DOI:10.1109/tce.2026.3714729
摘要
The integration of intelligent consumer electronics into FemTech offers unprecedented opportunities for personalized oncology. However, deploying multimodal wearable systems to monitor treatment responses in specific molecular subtypes, such as RAS/RAF-wildtype colorectal cancer (CRC) in women, requires stringent privacy safeguards and clinical interpretability on edge devices. This paper presents ExplainCRC, an edge-deployable, privacy-preserving consumer wearable framework tailored for the efficacy stratification of female patients with RAS/RAF-wildtype CRC. ExplainCRC fuses consumer-grade sensor streams—photoplethysmography (PPG), electrodermal activity (EDA), skin temperature, and acceleration—into a unified representation using a multilayered attention encoder network. To establish trust with clinical experts and female patients, we employ layer-wise relevance propagation (LRP) to quantify the temporal contributions of each sensor stream to health outcome predictions. Addressing the critical privacy demands of women’s health data, the system leverages federated learning secured by a Gaussian differential privacy mechanism (ϵ = 2.8, δ = 10−5). Furthermore, we developed a mixed-precision model compression pipeline optimizing ExplainCRC for seamless execution on mobile platforms. Extensive evaluations across three public datasets—WESAD, PPG-DaLiA, and TCGA-COAD (utilizing genomic features as static surrogates for outcomes in a 3-category scheme)—demonstrate robust performance. Explain-CRC achieves a high explanation fidelity (0.867 and 0.843 on the two wearable benchmarks; 0.863 averaged across all three), and its throughput on a mid-range mobile platform reaches 21.4 frames per second for the compressed model (12.8 FPS for the full model), a 1.67× latency reduction relative to its own uncompressed counterpart. These technical results motivate, but do not yet establish, a path toward home-based monitoring of targeted-therapy responses, which will require dedicated clinical data collection and validation.
科研通智能强力驱动
Strongly Powered by AbleSci AI