Demetics Launches AI-Powered Breast Cancer Diagnostic Solution Leveraging Ultrasound Intelligence to Transform Medical Imaging

Breast cancer has become one of the malignant tumors that seriously threaten the health of women in China. On March 23, 2018, the National Cancer Center released the latest data on breast cancer among Chinese women, estimating the incidence and mortality of breast cancer in Chinese women in 2014 (due to the time required for data collection and statistical analysis, there is generally a three-year lag in data availability). In 2014, there were approximately 278,900 new cases of female breast cancer nationwide, accounting for 16.51% of all malignant tumor incidences among women, ranking first among malignant tumors in women.

Early diagnosis and early treatment are the keys to breast cancer management. Currently, only 20% to 25% of breast cancer cases among Chinese women are detected at Stage I, resulting in a five-year survival rate of approximately 80%. However, if breast cancer is identified in its early stages and patients receive standardized treatment promptly, the five-year disease-free survival rate can reach 95%, and the five-year survival rate for patients with Stage II disease can exceed 80%.

Faced with enormous screening demands, the scarcity of physicians and the misdiagnoses and missed diagnoses associated with manual image interpretation have become critical pain points requiring urgent resolution. Currently, artificial intelligence (AI) has emerged as a technological solution to address physician efficiency challenges. Recently, at the 2018 Top 100 Future Healthcare Companies Forum, Deshang Yunxing launched an AI-assisted diagnostic product for breast cancer. This marks another addition to its portfolio of medical auxiliary diagnostic products, following its 3D visualization-based precise preoperative planning, intraoperative navigation, postoperative assessment, and intelligent AI-assisted diagnostic system for thyroid nodules via ultrasound.

In response to the high incidence of breast cancer, commonly used non-invasive adjunctive screening modalities include mammography, ultrasound, and magnetic resonance imaging (MRI).

Among the three screening modalities, mammography is cost-effective and primarily utilizes X-ray imaging. Both domestically and internationally, it has become the primary method for breast cancer screening in women. This approach offers the greatest advantage in detecting microcalcifications, thereby enabling the identification of asymptomatic or non-palpable tumors, with diagnostic efficiency even surpassing that of magnetic resonance imaging (MRI).

The advantages of ultrasound lie in the fact that breast ultrasound involves no ionizing radiation and can be repeated as needed. Furthermore, ultrasound clearly delineates tissue layers and achieves 100% accuracy in differentiating cystic lesions (fluid-filled nodules) from solid masses, allowing for a preliminary assessment of tumor benignity or malignancy. Ultrasound can guide needle biopsy and evaluate axillary and supraclavicular lymph nodes for metastasis. Additionally, it is convenient and cost-effective, offering diagnostic advantages for dense breast tissue and mammary gland hyperplasia.

Another non-invasive diagnostic modality is magnetic resonance imaging (MRI). This type of examination is characterized by high sensitivity and absence of ionizing radiation hazards; however, due to its high cost, it is not suitable for population-based screening. It is generally employed as a further diagnostic tool for patients whose conditions remain inconclusive after mammography and color Doppler ultrasound.

In China, breast cancer screening currently targets women aged 30 to 60 through various methods. Unlike Western women, who tend to have more fatty breast tissue, over 50% of Chinese women have dense breasts with less fat. This requires radiologists to carefully scrutinize imaging studies for tiny masses, calcifications, or architectural distortions hidden within the dense tissue, a process that is typically time-consuming.

Compared with their counterparts in Europe and the United States, doctors in China face a significantly heavier workload in interpreting medical images. Zhang Juan, head of breast imaging products at Deshang Yunxing and a senior medical imaging expert with over 30 years of experience in diagnostic imaging, told reporters that while doctors in Europe and the United States may interpret images for approximately 20–30 patients per day, their Chinese counterparts need to review scans for at least 50 patients daily. Writing reports from morning till night often leads to fatigue.

Faced with immense demand, there is a severe shortage of physicians specializing in breast diagnostics, and the lengthy training cycle has become a significant challenge for radiology departments. From medical education through hospital-based standardized residency training, cultivating a senior expert in breast cancer diagnosis requires at least 5 to 10 years.

Given the high incidence of breast cancer, earlier detection leads to better treatment outcomes. Artificial intelligence is the most suitable auxiliary tool to liberate physicians from the burdensome task of image interpretation and to reduce misdiagnoses and missed diagnoses.

Deshang Yunxing Medical Technology, established in 2013, specializes in ultrasound-based artificial intelligence, applying AI technologies to preoperative planning, intraoperative navigation, and postoperative assessment for tumor interventional surgeries. The company’s R&D team, anchored by renowned experts and scholars, is led primarily by young Ph.D. holders who have conducted academic research at prestigious international institutions such as Harvard University, the Courant Institute of Mathematical Sciences at New York University, and the Einstein Institute in Germany. They have published nearly 200 academic papers in authoritative international journals.

Hu Hairong, Chairman and General Manager of Deshang Yunxing, stated, “We are not merely developing AI-assisted diagnostic systems for medical imaging; rather, we integrate assisted diagnosis with ultimate therapeutic applications to build an AI-powered precision diagnosis and treatment platform.”

In this newly launched AI-powered breast cancer auxiliary diagnostic product, Deshang Yunxing leverages ultrasound and X-ray imaging modalities, augmented by artificial intelligence technologies. The entire R&D process involved cleaning and organizing massive datasets. Senior breast specialists strictly adhered to the international ACR standards to personally annotate and extract breast imaging features, with repeated verification to ensure high accuracy.

In terms of product functionality, Deshang Yunxing performs qualitative and quantitative analysis of lesions in accordance with the Chinese Guidelines for Breast Cancer, and classifies lesions using the BI-RADS system based on ACR standards. Meanwhile, tailored versions will be developed for tertiary hospitals and primary care facilities to align with China’s specific healthcare context. This approach aims to alleviate the burden on large hospitals, provide auxiliary training for grassroots medical institutions, and truly facilitate the implementation of tiered diagnosis and treatment.

In certain specific tasks, such as facial recognition and other image-related applications in the field of computer vision, computers have already surpassed the human eye. The essence of artificial intelligence lies in the fact that as data volume increases, processing speed accelerates and costs decrease. This engineering progress has led to an inflection point in practical applications. In the healthcare sector, AI-assisted imaging diagnosis technologies based on big data and artificial intelligence have become a research hotspot. Hospitals and physicians are increasingly embracing AI, delegating tedious yet straightforward preliminary tasks to artificial intelligence systems.

In the application of ultrasound AI, Shanghai Ruijin Hospital initiated and convened more than 400 hospitals across China in August 2018, bringing together nearly 1,000 sonographers to establish the “China Thyroid and Breast Ultrasound Artificial Intelligence Alliance.” This alliance built a national thyroid and breast ultrasound database covering multi-tier medical institutions throughout China, with ultrasound AI providing effective support for precise diagnosis and treatment. Unlike the crowded field of radiology AI products, the ultrasound AI sector is not yet saturated, with relatively few companies having established a presence. Consequently, this domain presents significant opportunities.

Ultrasound is widely available in primary care hospitals due to its low cost and lack of radiation; however, there is a severe shortage of sonographers. Unlike CT, MRI, and X-ray imaging, where image acquisition and interpretation are separate processes, ultrasound requires simultaneous image acquisition and interpretation. Furthermore, unlike MRI, CT, and ECG results, ultrasound diagnosis relies largely on dynamic images acquired from various planes by the operator. This places high demands on the sonographer’s technical skills. Variations in scanning techniques, individual patient differences, and inter-observer variability among physicians can easily lead to misdiagnosis or missed diagnoses. Deshang Yunxing has developed an AI-assisted diagnostic product for breast cancer, aiming to facilitate tiered diagnosis and treatment and alleviate key operational challenges faced by physicians.

Although the development of ultrasound AI still faces certain challenges, such as a lack of large-scale training data, major companies remain enthusiastic about this field. In addition to Tencent, industry giants like Samsung, Siemens, and Mindray have also made strategic moves in ultrasound AI. Furthermore, startups such as Infervision and Yitu Healthcare have launched relevant ultrasound AI products.

Hu Hairong believes that the advantages of Deshang Yunxing’s AI-assisted diagnostic products stem from two aspects: data and algorithms.

In terms of data, the training dataset comprises tens of thousands of cases. Regarding the quality of data annotation, Deshang Yunxing has collaborated with multiple hospitals, and the annotations were personally performed by a breast specialist with over 30 years of experience. She emphasized, “Data annotated by experts yields significantly higher accuracy in lesion detection compared to annotations performed by general physicians.”

Currently, this AI-powered diagnostic product for breast cancer achieves an accuracy rate of 89% in mass detection and over 90% in calcification identification. With continuous algorithm optimization, its accuracy is steadily improving. When deployed in clinical settings, the diagnostic performance reaches the level of senior expert physicians (chief physicians).

In terms of algorithms, it is reported that the algorithm team at Deshang Yunxing has a background in applied mathematics. Therefore, they independently developed a deep learning framework named Light3, signifying “lightweight and flexible.” For different diseases, the framework and parameters can be adjusted as needed to improve model accuracy. In contrast, open-source deep learning frameworks cannot be arbitrarily modified, thus limiting the extent to which accuracy can be improved. Hu Hairong stated, “We have conducted our own comparisons; in the thyroid ultrasound project, Light3’s accuracy is approximately 30–40% higher than that of open-source alternatives.”

In terms of product design, Deshang Yunxing has developed differentiated solutions tailored to the needs of tiered diagnosis and treatment. For tertiary hospitals (Grade 3A), the product offers comprehensive functionalities to meet their research requirements. For primary care institutions, the system adopts a “streamlined” approach by removing redundant features to better address their basic clinical needs.

Furthermore, the product is tailored to hospitals at various levels and can be used to train junior physicians. Given the widespread shortage of medical professionals, it is not feasible to send these doctors back to academic institutions for specialized training; therefore, leveraging AI-assisted diagnosis serves as an effective training modality.

Operating in a relatively uncrowded niche, Deshang Yunxing’s ultrasound AI products are exploring their own business model. For most AI companies, obtaining relevant regulatory certifications represents a significant hurdle. Yan Yeen, General Manager of Marketing at Deshang Yunxing, stated that while current medical AI products on the market are predominantly concentrated in the radiology sector, Deshang Yunxing’s AI solutions hold comprehensive leading advantages in the ultrasound domain. Due to regulatory constraints, the AI industry as a whole has yet to develop mature business models; this limitation, in turn, raises the entry barriers for medical artificial intelligence products.

He emphasized, “Desyun Xing will increase its market investments, including seeking partners across the upstream and downstream value chains. This includes not only tertiary Grade-A hospitals but also primary care hospitals with substantial experience in product lines that can assist with data R&D and product promotion. Currently, our primary focus remains on enhancing product quality and accuracy. Once regulatory barriers are lifted, we will leverage our competitive advantages to carve out a differentiated path in the overall sales market.”

Currently, Deshang Yunxing has introduced industrial capital from Jia Shares in its Series A financing, and funds from Fosun Pharma and Huagai Capital in its Series B financing. According to Hu Hairong, Deshang Yunxing is currently securing cooperation for the next round of funding and resources, aiming to partner with investors and collaborators who possess extensive medical resources within the industry, to truly implement artificial intelligence products in clinical diagnosis and treatment.