HUST Research Applies Artificial Intelligence to Medical Diagnosis

In the context of the country strongly transforming on the journey of innovation, startup stories from university lecture halls are becoming practical solutions, creating profound value for society. Participating in the 8th National Startup Festival for Students, the name VietMedX³ shone, winning the Championship in the Industry - AI - STEM category, as a vivid testament to the creative startup spirit of Hanoi University of Science and Technology students.

Solving the Medical Imaging Problem with AI

In the current healthcare system, reading and analyzing PET/CT images - an advanced medical imaging technique that helps detect cancer and many dangerous diseases early - requires high professional qualifications and significant processing time. A diagnosis can take tens of minutes, or even longer if the data is complex. Meanwhile, the increasing number of patients keeps doctors in a constant state of overload.

And that was the opening question for the research of HUST students: "Can AI share this burden?".

VietMedX³ emerged as a promising solution. With the ability to process data quickly, the system can shorten the diagnosis time to just 5-15 minutes, while automatically generating detailed reports, supporting doctors in making more accurate decisions.
  The system is built on an advanced artificial intelligence model, going through three main training phases to create the final product, VietMedX³ - a model capable of automatically generating medical reports and answering questions.

First is the representation learning phase, where a 3D encoder is trained using self-supervised learning to extract features from millions of PET/CT images, helping the AI understand human body structure through input images.

Next is the alignment phase, in which the model is frozen, and only the 3D projector is trained to translate image data into language. Finally, the fine-tuning phase, where the 3D projector and the language model applying LoRA are trained to execute clinical commands.

Notably, VietMedX³ was trained by the research team on the ViMed-PET dataset consisting of over 1.5 million whole-body PET/CT images from 2,757 actual clinical cases.

And the test results speak for themselves: VietMedX³ achieved an accuracy of up to 90.55% in trials, an impressive figure for a newly developed AI system, outperforming GPT-4o (81.09%), RadFM (69.49%), and many other medical AI models.

Sharing about the difficulties during the project implementation, team leader Nguyen Hai Dang expressed: "Despite being confident about our technological strengths, when putting ourselves in the context of a startup, the team had to face many finance-related issues, especially in the commercialization phase."

To solve this problem, the members proactively learned more about economics and investment, while seeking opportunities to meet and work with experienced mentors to gradually perfect the product completely. That journey of overcoming challenges was also the glue that bound the members together, creating a collective strong in expertise, flexible in thinking, and persistent in pursuing goals.

The Team Behind VietMedX³

Behind VietMedX³ is a young collective, each taking on a separate piece of the puzzle and operating as a unified whole. The team consists of 5 members, with four K68 students majoring in Data Science and Artificial Intelligence (IT-E10), School of Information and Communication Technology, Hanoi University of Science and Technology, and 1 female member from the National Economics University.

Among them, Nguyen Hai Dang is in charge of overall system architecture, core technology orientation, and R&D progress management; Chu Anh Duc is responsible for bringing the product to real-world environments at hospitals and medical facilities; Le Hoang Tung closely monitors the operation process and technical processing, ensuring the system operates stably; Do Dang Vu focuses on researching and training the 3D VLM model and optimizing the AI model on server hardware, while Tran Thi Nhu Ngoc is in charge of finance and connecting with users.

Member Le Hoang Tung shared about his role: "An AI system can be very strong algorithmically, but if it doesn't operate stably, it's hard to create value in reality. My job is to ensure everything always runs smoothly, so doctors can trust using it every day."

Accompanying the team are two advisors holding orientation roles: Assoc. Prof. Nguyen Phi Le - Executive Director of the Artificial Intelligence Research and Application Institute (AI4LIFE) - is the one guiding academia and research; CETECH Chairman Tran Trong Viet is the one bringing market perspectives and the commercialization path, taking on the role of financial advisor for the research team.

With an entrepreneurial spirit and smooth coordination among members and the advisory team, VietMedX³ has gradually perfected from a research idea into a highly applicable solution. The project made its mark when it previously won the Second Runner-up prize at the AI+X Global Startup Competition, Sungkyunkwan University, South Korea - a country considered one of the leading tech hubs. This achievement shows that VietMedX³ is fully capable of stepping into the international arena with remarkable competitiveness.
  After winning the Championship at the 8th National Startup Festival for Students, VietMedX³ plans to continue implementing its "go to market" strategy, aiming for commercialization and deployment at hospitals nationwide. If widely applied, this solution could create revolutionary changes in the healthcare system.

In a rapidly changing world where technology has become the main driver of development, initiatives like VietMedX³ are the key to promoting Vietnam's science and technology.

Imagine a future where every doctor has an "AI assistant" to support their daily work. Diagnosis time is shortened, errors are minimized, and patients receive results faster. That would be a leap forward in the quality of healthcare, originating from a research project at Hanoi University of Science and Technology.

Who knows, in the not-too-distant future, when mentioning the advances of smart medicine, people will remember VietMedX³ - a "made in Vietnam" product, created from the intellect, passion, and aspiration to contribute of the young generation of Hanoi University of Science and Technology!