This article is shared from Huawei Cloud Community "[ Pioneer Developer Cloud Talk] After four years of hard work, a group of young "tumor recognizers" use AI to help medical innovation ", author: Gauss Squirrel Club Assistant 2.
A group of young "tumor recognizers" have worked hard for four years to overcome smart medical problems and ignite the spark of the health revolution!
"Good news" for patients with skin tumors
Yang Yang, a junior majoring in embedded software at Zhengzhou University of Light Industry, is currently the third-generation leader of the Tumor Finder project at Mekor Studio.
Maker Studio was established at Zhengzhou University of Light Industry in 2014. Its research interests cover more than 10 fields such as medical devices, 3D printing, soft robots, and artificial intelligence. More than 3,000 students from different majors have joined. Taking the medical industry as an example, the studio has accumulated more than 300 application cases, and the tumor recognizer project is one of them.
The Tumor Recognizer project focuses on the research and development of malignant skin tumor identification technology to assist doctors in quickly and accurately analyzing and diagnosing through medical images in a short time. In this project, in addition to Yang Yang, who is responsible for hardware development, team members also include Chen Yilin, who is responsible for QT software development, Chen Yuling, who is responsible for UI design, Chen Xinjie, who is responsible for algorithm development, and Xie Zhengru, who is developing HarmonyOS.
Since its launch in 2020, the project has gone through many iterations and upgrades. Currently, six types of malignant skin tumors, including melanoma and basal cell carcinoma, have been identified. In clinical testing, joint doctors conducted actual tests on 30 patients and detected 11 cases of melanoma and 3 cases of benign keratosis, with an accuracy of 91.6%.
Click me to view the full video
With the blessing of Huawei Cloud, "tumor recognizers" continue to iterate and upgrade
Perhaps it comes from the studio's research and analysis of current industry information technology. After understanding and applying the collaborative capabilities of Huawei's diverse ecosystem, the studio migrated all cloud services to Huawei Cloud during the summer of 2020. Subsequently, the studio and Huawei Cloud began to cooperate closely in terms of software and hardware.
With the support of Huawei Cloud, "tumor recognizers" have achieved a leap from theory to practice. Starting in 2020, the Tumor Recognizer Project has used Huawei Cloud AI development platform ModelArts to build and optimize algorithms. Through the unremitting efforts of the team, the project was iteratively upgraded this year and a software and hardware architecture solution based on Ascend AI and Huawei Cloud ModelArts was launched.
In order to identify tumors more accurately, Yang Yang's team collected patient data through dermoscopy and then uploaded it to Huawei Cloud OBS object storage service for directional data storage. This cloud data processing method not only improves the security of data storage, but also provides strong support for subsequent research and development work.
Throughout the project, Yang Yang's team used Huawei Cloud ModelArts as the AI application development platform for the entire project to perform data processing, model training, and AI application management. Among them, in terms of data processing, they combined the geometric transformation and color transformation features of ModelArts to enhance the data, and ultimately improved the accuracy of the model by increasing the diversity of the data set. During the development process, the actual experience of ModelArts' automatic learning service and model training and deployment capabilities inspired Yang Yang's team to further explore AI technology.
In terms of algorithm construction and optimization, Yang Yang and his team demonstrated their in-depth understanding and application capabilities of technology. Using Huawei's open source AI framework MindSpore, they successfully built the ResNet50 malignant skin tumor classification algorithm. After many optimizations, it not only improved the accuracy of the model, but also reduced the amount of code by 30%, which is of great significance for the subsequent development and maintenance of the project.
It must be mentioned that in the model optimization process, the MindSpore visual debugging and tuning tool MindinSight in the Ascend application enabling platform MindX played a key role. It helped the team complete model traceability and hyperparameter search, allowing them to find the best model parameters in a short time, thereby improving the model's F1-Score to 0.908. ( F1-Score is a weighted average of model accuracy and recall. The maximum value is 1 and the minimum value is 0. The larger the value, the better the model )
In terms of end-side development, Yang Yang and the others also tried to deploy this algorithm on the edge-side device Ascend Atlas 200I DK A2 developer kit, and finally achieved millisecond-level image inference recognition locally, thus significantly shortening the entire detection and diagnosis time to 30s ~ 1m.
In order to enable doctors and patients to use it better, the team not only developed a Windows-based doctor client web page, but also developed a Hongmeng APP based on HarmonyOS to facilitate patients to upload images and query diagnosis and treatment records.
Talking about the experience of cooperation with Huawei Cloud in the "Tumor Recognizer" project, "In addition to the diversified support of technical products, Huawei Cloud provided us with a lot of technical and resource support during our actual application development process. From answering questions about the product at the beginning, to all kinds of difficult problems in the current project, Huawei Cloud's technical experts will explain everything." Yang Yang said happily, "We have already discussed with Huawei Cloud's technical experts Many experts have become great friends.”
Not only in terms of technical support, but also in terms of AI model training, the team also received a 20,000 yuan voucher and a Shengteng 910 training card provided by the Zhongyuan Artificial Intelligence Computing Center. In terms of personal growth, team members often participate in technical training, developer competitions and other activities organized by Huawei Cloud, using competitions to promote learning so that they can truly master technical knowledge in practice.
Up to now, Yang Yang has led the "Tumor Recognizer" team and has won many competition awards such as the third prize in the Shenzhen Division of the 2023 Huawei Developer Competition, the third prize in the Henan Regional Developer Kit Innovation Track of the Ascend AI Innovation Competition 2023.
Break cognitive limitations and achieve technological breakthroughs and growth gains
"The five of us gained a lot during the project." Yang Yang said that for them in the Central Plains region, many resources are limited. By participating in these projects at Mekor Studio, they can gain real Practical development capabilities and experience in scenarios.
It was these projects that gave Yang Yang and the others the opportunity to come into contact with the actual development content of projects in a real production environment, and to learn about technologies beyond school textbooks, including cloud development, AI large model applications, databases that separate storage and calculation, etc., breaking through Overcome your own cognitive limitations and truly learn useful skills and knowledge at the university level.
Yang Yang gave an example. For example, he had done a lot of embedded software development before, so when he first came into contact with the tumor detector project, he was not very familiar with AI technologies such as deep learning. After that, I studied related free courses provided by Huawei Cloud, including ModelArts, Shengteng, MindSpore, etc., and then slowly got deeper into it.
Not only that, university developers like Yang Yang can experience and understand the learning paths and development case tutorials of many technologies and products from scratch in Huawei Cloud's developer community, cloud school, and other places. In order to better enable them to apply what they have learned, Huawei Cloud has also prepared training camps and developer competitions in different technical fields for developers, so that student developers can improve their capabilities and obtain more rewards.
"Being able to develop innovative applications based on cutting-edge technologies around real industry scenarios has allowed our student developers to truly appreciate the value of integrating industry and academia. We are also grateful to Huawei Cloud for providing diverse ecological technology and resource support to help us realize those wild imaginations. Embrace a more ambitious development blueprint." Yang Yang said.
at last
This is the journey of the "Tumor Recognizer" project team's "Chasing Dreams on the Cloud".
They use their own experiences to tell us that every young developer is a changer in the world, and every code and every idea they produce may become the key to changing the future. What supports them is platforms like Huawei Cloud. They use technology, resources, and feelings to help every youth realize their dreams.
With dreams in mind, young people are promising!
Click to follow and learn about Huawei Cloud’s new technologies as soon as possible~
Alibaba Cloud suffered a serious failure, affecting all products (has been restored). The Russian operating system Aurora OS 5.0, a new UI, was unveiled on Tumblr. Many Internet companies urgently recruited Hongmeng programmers . .NET 8 is officially GA, the latest LTS version UNIX time About to enter the 1.7 billion era (already entered) Xiaomi officially announced that Xiaomi Vela is fully open source, and the underlying kernel is .NET 8 on NuttX Linux. The independent size is reduced by 50%. FFmpeg 6.1 "Heaviside" is released. Microsoft launches a new "Windows App"