Breaking Away from 'One-Size-Fits-All': West China Hospital's AI-Powered Health Management System Delivers Truly Personalized Weight-Loss Plans

Image source: Official website of West China Hospital, Sichuan University

Recently, an invention patent of West China Hospital of Sichuan University is proposed to be transferred via an exclusive license, with the transferee being Chengdu Yunzuji Technology Co., Ltd.The transaction price is RMB 100,000 plus a 2% royalty on sales revenue.. Patent name:"A Health Management Method and System Based on Multi-Source Data", inventors includeChen Yi, Huang Jianbo, Feng Feiyun, Li Yanju, Chen Haiyang, Wan Qianyi, and Zhao Jie。

The core of this patent isA Full-Loop Health Management System: "Data Integration → Solution Generation → Compliance Tracking → Dynamic Optimization", consisting of three technical modules:

Multimodal Data Fusion Layer.Traditional health management software relies on hospital medical records and test reports, resulting in single-dimensional data with lagging updates. This patent integrates five major categories of information: exercise data, sleep data, weight and body fat metrics, lifestyle habits, and physical examination reports, covering comprehensive dimensions both outside and inside hospitals, as well as static and dynamic data. After denoising and feature extraction for each data type, a unified health profile is generated through fusion algorithms—this serves as the data foundation of the entire system and the basis for subsequent solution generation.

Fine-tuning Large Language Models with Medical Knowledge Bases.Leveraging open-source pretrained large language models such as Llama and Qwen, we constructed a proprietary medical knowledge base using West China Hospital’s clinical knowledge graph, medical diagnostic reports, and fat-loss regimen library to fine-tune the models. Upon inputting user health characteristics, the model simultaneously generates medical diagnostic reports and personalized fat-loss plans. These plans are tailored to the user’s actual metabolic levels and daily routines, significantly enhancing feasibility—thereby addressing the core limitation of traditional fat-loss products, which offer a “one-size-fits-all” approach.

Adherence Closed-Loop and Incremental Training.After users implement the plan, the system continuously collects metric data such as weight, exercise adherence rate, and body fat improvement. It categorizes samples into compliant and non-compliant groups to iteratively optimize the model. Meanwhile, gradient low-rank projection technology is introduced to project the large model’s gradient matrix into a low-rank subspace, significantly reducing GPU memory usage and computational costs. This design enables small and medium-sized health management institutions to deploy AI-driven solution engines, making them accessible beyond just industry giants.

A closer look at the industry reveals that fragmented data, rigid solutions, and a lack of feedback are nearly universal ailments among all players.

Data Level，Most health management apps remain positioned as "recording tools", how many steps users take and how many hours they sleep each day—these data points remain siloed within devices, failing to be integrated into comprehensive health profiles. Wearable devices, smart body fat scales, and dietary habit logs—such out-of-hospital data have long remained outside mainstream healthcare systems.resulting in a lack of complete context for health analysis。

At the program level, most existing fat-loss and health management programs are generated based on preset rules or simple algorithms.For individuals with a BMI above the normal range, the advice received may be nearly identical, regardless of their metabolic rate or lifestyle habits. This "one-size-fits-all" approach overlooks the significant differences between individuals, inevitably compromising its effectiveness.

In terms of feedback, nearly all products stop at "providing solutions."Once the plan is delivered, did the user implement it? For how long? Did the metrics change? These questions remain unanswered, and there is no mechanism in place for follow-up inquiry and iteration. The plan becomes a “one-time delivery,” with no guarantee of long-term effectiveness.

The application of large language models in healthcare settings is not new, but their true implementation faces two hard thresholds:1. Medical Professionalism, large models suffer from "hallucination" issues, and the reliability of general-purpose model outputs is questionable in medical scenarios;Second, computational power costs, the resource consumption of a single full fine-tuning of a large language model is prohibitive for small and medium-sized institutions.

The solution proposed in this patent is:Fine-tuning with a medical knowledge base, rather than full-scale training, enhances professional expertise while controlling costs; employing gradient low-rank projection reduces computational demands, making incremental training feasible. This combined approach does not rely on the most advanced computational resources, representing a technical pathway that is readily accessible to small and medium-sized institutions.

Of course, the feasibility of a technical pathway does not guarantee smooth commercialization. The key variables determining the ultimate value of this patent are Cloud Footprint Technology’s implementation capabilities, channel resources, and user operation proficiency.

The combination of an exclusive license with sales royalties is not common in patent transfers. This indicates that West China Hospital sought not merely to "secure immediate gains," but rather chose to deeply align its interests with those of the licensee.

For a technology transfer affiliate of a hospital, this is a pragmatic and forward-looking arrangement: the RMB 100,000 minimum guarantee represents the floor, while the 2% sales royalty determines the ceiling. The stronger the commercialization capabilities of the licensee, the more substantial the long-term returns for West China Hospital. This alignment mechanism also compels the licensor to exercise greater diligence in screening potential licensees at the early stage; financial capacity alone is insufficient, as the licensee must also demonstrate the ability to successfully commercialize the technology.