A Grade-A Hospital: Building an IBS Intelligent Diagnosis System Based on MatrixOne Intelligence

A Grade-A Hospital: Building an IBS Intelligent Diagnosis System Based on MatrixOne Intelligence

Customer Profile

The customer is a Grade III Class A hospital in China, integrating medical services, education, and research. It possesses strong technical capabilities and advanced medical equipment. However, currently, in the clinical diagnosis and treatment of functional gastrointestinal disorders such as irritable bowel syndrome (IBS), it is facing inherent challenges and external pressures regarding the utilization of data assets.

Challenges

1. The inherent difficulties in clinical diagnosis and treatment.

IBS is a chronic disease, and in most cases, the symptoms do not pose a threat to life, so patients are often reluctant to go to the hospital for treatment. Compared to the complex offline consultation process, they prefer convenient online consultations. Meanwhile, the existing offline diagnosis model is not only time-consuming and labor-intensive, but also has the risk of incorrect diagnoses due to the lack of standardization in the consultation process. It is also difficult to provide long-term and personalized follow-up and management, which is significantly out of sync with the actual needs of IBS patients.

2. Difficulty in utilizing in-hospital data assets.

Hospitals have accumulated a large amount of valuable clinical data, but these data usually cannot be directly exported for analysis and model training. Data silo and non-standardization problem prevent valuable medical data from being effectively utilized, making it impossible to support large-scale data-driven research or the development of AI models, and limiting the deep integration of clinical research and practice.

3. External resource pressure and compliance challenges.

The large number of IBS patients has placed a huge pressure on the limited specialized outpatient resources, increasing the burden on the medical system. At the same time, as a medical device software, the intelligent consultation system must strictly follow regulatory requirements for its development and application, ensuring the safety and effectiveness of the product, and passing the clinical validation. This is the prerequisite for obtaining the registration certificate for Class II medical device software and launching it to the market, and is also a key aspect that review experts focus on.

Solution

To address these challenges, this hospital has introduced MatrixOne Intelligence — an AI data intelligence solution for multi-modal data — aiming to build an end-to-end IBS intelligent consultation system from data processing to intelligent applications.

The MatrixOne Intelligence platform integrates functions such as data governance, intelligent parsing, multi-modal search, and a hyper-converged data infrastructure. Through its core components, it can intelligently parse and extract structured data from various unstructured data types like PDFs, audio, video, images, and supports the construction and generation of high-quality training datasets based on specific AI model requirements, thereby providing a solid data foundation for the implementation of AI applications.

The specific implementation path is as follows:

1. Revitalize the data assets within the hospital

Utilize the capabilities of the MatrixOne Intelligence platform to conduct structured extraction of the identified unstructured text (after de-sensitization), converting it into standardized JSON format data according to the preset Schema (such as patient information, chief complaint, current medical history, diagnosis, etc.). This step successfully transforms the "dead data" originally locked in images into "live data" that can be analyzed and utilized by machines, laying a high-quality data foundation for the subsequent development of AI models.

2. Building a High-Quality AI Training Dataset

a. Basic Dialogue Generation: Based on the structured medical record data obtained in the previous step, a doctor-patient dialogue (QA) dataset that highly simulates real clinical scenarios was generated by leveraging locally deployed large language models (e.g., Qwen3-32B) and prompts designed in alignment with clinical consultation workflows.

b. Data Amplification and Optimization: To address the bottlenecks such as mechanical responses and "over-diagnosis" in the early stages of AI model development, the project has adopted advanced data amplification strategies:

• Introducing Persona characters: Simulating patient characters with different personalities and diverse expression styles, and regenerating dialogue data greatly enriches data diversity, enhances the model's empathy ability and the naturalness of the dialogue.

• Introducing negative Samples: To correct the overfitting tendency of the model to diagnose all conversations as IBS, the team introduced 10% to 20% of non-IBS cases or misleading adversarial samples as negative samples. This effectively trained the model's differential diagnostic ability, enabling it to more accurately distinguish IBS from other conditions.

3. Develop core applications for IBS intelligent diagnosis:

Based on high-quality datasets, the team successfully developed the IBS intelligent consultation system. This system is equipped with four core AI-assisted functions, deeply empowering clinical doctors:

• Intelligent Response: Understand the context of doctor-patient conversations, generate professional response suggestions in real time for doctors to choose from, and improve communication efficiency.

• Intelligent Diagnosis: Automatically extract key symptoms from conversations, conduct logical reasoning based on clinical guidelines, and provide diagnostic suggestions such as IBS subtype classification.

• Treatment Advice: Based on the diagnosis results and the patient's condition, a personalized comprehensive treatment plan including medication, diet, and lifestyle is recommended.

• Medical Record Report: One-click automatic integration of the entire consultation process information to generate standardized electronic medical record reports.

Customer Benefits

The implementation of this intelligent consultation system has brought significant multi-dimensional value to the clinical work, scientific research and patient services of this hospital:

● Enhance diagnosis and treatment efficiency: The automated consultation and medical record generation functions free doctors from repetitive work, saving a great deal of time.

● Standardize diagnosis and treatment procedures and patient management: The system ensures that data collection adheres to the latest clinical guidelines and expert consensus, reducing the risk of missed diagnoses. Meanwhile, the system can generate personalized evidence-based treatment plans based on the specific conditions of patients, which is conducive to improving patients' treatment compliance and long-term management effects.

● Optimize the allocation of medical resources and patient experience: Through online pre-consultation, patients with mild symptoms can be effectively diverted, alleviating the reception pressure on the specialized outpatient department of gastroenterology. Patients have also gained more convenient and private online consultation channels, and their overall medical experience and trust have been enhanced.

● Promoting the development of clinical research: This project successfully transformed hospital's data assets into valuable research resources. The large amount of structured real-world data accumulated by the system provides unprecedented data support for exploring the pathogenesis of IBS, subtype conversion rules and other cutting-edge scientific research.