VoxelCloud Finds a New Commercial Path in Internet Healthcare Empowerment [2021 WAIC]

Since IBM Watson announced its ambition to become a physician in 2011, the field of medical artificial intelligence has experienced a decade of ups and downs.

After years of trial and error, the first generation of single-disease AI-powered medical imaging devices, spun out from Grade A tertiary hospitals, has reached sufficient technological maturity. The future development will focus on two directions: one is to expand from radiology departments to clinical specialties, thereby broadening the scope of AI applications; the other is to establish a strong presence at the primary care level, integrating AI with 5G and Internet of Things (IoT) technologies to maximize the value of artificial intelligence in improving efficiency and reducing costs.

Regardless of the chosen path, medical AI companies must continuously refine their products, transitioning from single-disease focus to comprehensive disease coverage. The operational logic of imaging AI should shift from “judging predefined outcomes (such as pulmonary nodules)” to “one scan, full-spectrum disease insight,” thereby realizing the true essence of “preventive medicine at its highest level.”

Since its inception, VoxelCloud has recognized this trend and focused on the development of artificial intelligence (AI) solutions for a wide range of diseases. To date, VoxelCloud has completed AI development for three clinical scenarios: chest CT, fundus imaging, and dermatology. At the 2021 World Artificial Intelligence Conference (hereinafter referred to as “WAIC”), Zhang Zheng, Chief Operating Officer of VoxelCloud, delivered a speech at the Health Summit Forum, redefining the future development path of AI in medical imaging.

Artificial intelligence (AI) refers to human-like intelligent programs constructed using algorithms, data, and knowledge, renowned for improving efficiency and reducing costs. Therefore, against the backdrop of uneven distribution of medical resources and a lack of awareness in disease prevention and control, primary healthcare represents the largest application scenario for AI. After all, the shortage of manpower and lack of hardware equipment in primary healthcare cannot be readily addressed in the short term. By integrating with communication technologies, AI may serve as the most effective lever for implementing tiered diagnosis and treatment.

In its exploration of primary healthcare, VoxelCloud has primarily focused on two directions: one is an AI-based solution for fundus diseases, and the other is “VoxelSkin,” an internet consultation-based AI solution for dermatology.

The AI exploration of fundus diseases stems from the vast patient demand in China. “China has 114 million people with diabetes and 400 million middle-aged and elderly individuals aged 50 and above, all of whom are at high risk for ocular diseases,” explained Zhang Zheng. “Screening via fundus photography is the most effective method to prevent or reduce blindness. However, China has fewer than 40,000 ophthalmologists in total. Data show that the annual number of outpatient visits for fundus-related diseases is approximately 110 million. Based on these figures, the screening rate for fundus diseases in China is less than 10%.”

To address this need, VoxelCloud has designed an integrated hardware-software AI solution for fundus disease screening. This solution combines traditional devices, such as handheld slit lamps and fully automated handheld fundus cameras, with algorithms independently developed by VoxelCloud. Empowered by AI, the system enables technicians without a medical background to capture images with a single click and receive AI screening report results within two seconds of image upload.

Zhang Zheng told VCBeat, “VoxelCloud has partnered with the China Chronic Disease Management Center to provide AI-powered screening services to 260 hospitals under its network. Over 200,000 diabetic patients have already benefited from these AI-based screening services. Data collected during the collaboration period show that the concordance rate between VoxelCloud’s AI results and physicians’ manual assessments exceeded 86%.”

“With high accuracy, this technology can generate economic benefits for hospitals. Taking Tianjin First Central Hospital as an example, while we assisted the Department of Endocrinology in conducting fundus screening, we also helped the hospital’s Ophthalmology Department retain more patients for surgical procedures. Rough estimates indicate that, empowered by AI, the hospital’s monthly revenue could increase by approximately RMB 500,000.”

To date, VoxelCloud’s AI-assisted fundus screening initiatives have been primarily implemented in densely populated provinces with relatively robust county-level economies, including Hebei, Henan, Shandong, and Jiangsu. In this process, VoxelCloud collaborated with the Hebei Provincial Eye Hospital and Hebei grassroots health centers to establish the Hebei Provincial Grassroots Ophthalmic Prevention and Treatment Alliance, effectively enhancing the coverage and efficiency of fundus disease screening among residents.

Zhang Zheng stated, “On one hand, we collaborate with provincial-level hospitals to develop AI products, which are then deployed at the village or township level. If residents complete screening using our devices, the AI can identify patients requiring surgery and recommend their referral to higher-level hospitals. This approach will effectively facilitate the advancement of tiered diagnosis and treatment.”

Unlike AI solutions for fundus disease screening, VoxelCloud’s dermatology AI product, “VoxelDerm Insight,” is a typical B2B2C offering. Its application scenarios stem from the rapidly growing field of internet-based healthcare in recent years.

For most diseases, doctors struggle to obtain sufficient patient information due to limited physical interaction; consequently, internet-based healthcare typically offers only preliminary consultation services, providing recommendations for subsequent diagnosis and treatment based on the information supplied by patients. In contrast, online dermatology consultations possess inherent advantages. After all, dermatologists primarily rely on visual data for diagnosis, and through the internet, patients can easily provide high-quality images of their skin conditions to physicians.

Specifically, VoxelCloud SkinAI can provide intelligent auxiliary diagnostic services while users upload photos, without altering the existing clinical workflow. Additionally, patients can perform self-examinations at home via the VoxelCloud SkinAI WeChat Mini Program, which helps them identify potential issues independently or enhances post-treatment follow-up.

After four years of refinement, VoxelCloud Skin AI has entered its 2.0 version. At this stage, the AI can diagnose 143 types of skin diseases, covering more than 99% of skin disease cases. The model’s diagnostic accuracy rates for top-1, top-3, and top-5 predictions are 71%, 89%, and 94%, respectively. In comparison, the accuracy rate for diagnosing the benign or malignant nature of skin tumors at primary-care hospitals is only about 30%, while the average accuracy rate among physicians at tertiary Grade A hospitals is approximately 70%. This means that the upgraded “VoxelCloud Skin AI” product now matches the disease coverage and diagnostic accuracy of dermatologists at tertiary Grade A hospitals.

Data provided by VoxelCloud shows that the diagnostic accuracy of general practitioners in dermatology is merely 24%-70%, whereas the top-5 diagnostic accuracy of VoxelCloud’s AI-powered multi-disease skin detection product reaches as high as 93%. With the assistance of VoxelCloud’s Skin AI, it is expected that the diagnostic and treatment capabilities of primary-care general practitioners will rival those of dermatologists (whose diagnostic accuracy ranges from 77% to 96%), thereby truly achieving the goal of strengthening primary healthcare.

As consistently emphasized by all artificial intelligence companies, VoxelCloud has developed high-accuracy AI not to replace dermatologists, but rather to help them eliminate the cumbersome, low-value repetitive tasks in their daily clinical practice, thereby allowing them more time to focus on and invest in high-value activities, such as managing common and frequently occurring conditions like psoriasis, urticaria, and acne.

To establish a comprehensive disease screening and prevention system, we need more than just accurate diagnoses for individual patients. More importantly, we must engage physicians in retrospective disease studies, the exploration of novel therapeutic approaches, and health education initiatives, thereby enabling more patients to receive treatment at the early stages of disease. This is the original mission behind VoxelCloud’s Design AI.

Generally speaking, medical AI companies regard tertiary hospitals as payers, selling their AI solutions to these hospitals through bidding processes to generate revenue.

However, within VoxelCloud’s commercialization framework, tertiary hospitals primarily function as partners. Specifically, these hospitals collaborate with the company to co-develop artificial intelligence solutions and deploy them at the primary care level. Through this process, VoxelCloud generates revenue from AI sales, while the hospitals enhance their capacity to retain patients for follow-up care.

Furthermore, the B2B sector is also a crucial pathway for VoxelCloud to achieve commercialization. By integrating AI into the online consultation processes of internet healthcare companies, platforms can leverage VoxelCloud’s FuZhiHui system to provide rapid and accurate diagnostic results for patients who upload dermatological images. This will effectively enhance the value of internet healthcare services, thereby increasing user stickiness.

“During the pandemic last year, our AI dermatology product was featured in WeChat’s nine-grid service hub, garnering over one million visits within less than a month. At that time, many patients found it inconvenient to visit hospitals, but they were able to receive timely medical consultations using this tool,” said Zhang Zheng.

“In addition to dermatological conditions, VoxelCloud Skin Insight can also be applied in scenarios such as medical aesthetics and beauty skincare for timely detection. For instance, the skin camera we integrated into Alibaba’s Quark platform has become the most frequently used feature for online consultations on that platform. Winona serves as a prime example of our business development in the beauty and skincare sector. The company leverages daily skin management for its members to enhance user engagement. The introduction of VoxelCloud Skin Insight has significantly improved the accuracy of this management process while reducing costs, enabling Winona to manage a large volume of customers on a daily basis at a low cost.”

Currently, VoxelCloud has established partnerships with Tencent Health, AliHealth, Miaoshou Doctor, Winona, and other enterprises. As of July 2021, the number of API calls for VoxelCloud’s AI-assisted dermatological diagnosis service had exceeded 30 million, serving numerous B-end clients.

However, in VoxelCloud’s view, 30 million is a respectable figure, but merely a number. The value of AI lies in amplifying the value of data to better serve physicians, benefit patients, and advance the healthcare system. To achieve this goal, VoxelCloud still has a long and arduous journey ahead.