How a Student Team Achieved 91.6% Accuracy in Skin Cancer Detection with Huawei Cloud AI

01

Yang Yang, a third‑year embedded software student at Zhengzhou University of Light Industry, leads the third generation of the "Tumor Identifier" project in the Maker studio, which was founded in 2014 and now includes over 3,000 students working on medical devices, 3D printing, soft robotics, AI and more.

The project focuses on malignant skin tumor identification, helping doctors quickly and accurately diagnose from medical images. The team consists of hardware developer Yang, QT software developer Chen Yilin, UI designer Chen Yuling, algorithm developer Chen Xinjie, and HarmonyOS developer Xie Zhengru.

Since its launch in 2020, the system can recognize six malignant skin tumors, including melanoma and basal cell carcinoma, achieving a clinical test accuracy of 91.6% on 30 patients (11 melanomas, 3 benign lesions).

02

In 2020 the team migrated all services to Huawei Cloud, using ModelArts for algorithm construction and optimization. Patient skin‑lesion images are captured with a dermatoscope, uploaded to OBS object storage, and processed with ModelArts' geometric and color transformations for data augmentation.

Using Huawei’s open‑source AI framework MindSpore, the team built a ResNet‑50 classifier for malignant skin tumor classification, reducing code volume by 30% while improving precision. The MindX visual debugging tool MindInsight was employed for model tracing and hyper‑parameter search, raising the F1‑Score to 0.908.

For edge deployment, the model was ported to the Ascend Atlas 200I DK A2 development kit, achieving millisecond‑level inference and reducing total diagnosis time to 30 seconds–1 minute.

03

The project also produced a Windows‑based doctor client web page and a HarmonyOS app for patients to upload images and view diagnostic records. Team members received extensive support from Huawei Cloud experts, training resources, and a ¥20,000 voucher plus Ascend 910 training cards from the Central Plains AI Computing Center.

Through participation in Huawei Cloud’s developer community, training camps, and competitions, the team won the 2023 Huawei Developer Competition Shenzhen regional third prize and the Ascend AI Innovation Competition Henan regional third prize.

04

The "Tumor Identifier" journey demonstrates how young developers can combine cloud services, AI frameworks, and edge hardware to create impactful medical applications, turning academic projects into real‑world solutions.