SenseTime Launches Two MICCAI 2019 Grand Challenges in Pathology and Radiotherapy to Accelerate AI Clinical Adoption

As a Gold Sponsor of MICCAI, the premier international conference on medical image computing, Sensetime recently announced that it would join forces with Hengdao Pathology, Ruijin Hospital affiliated to Shanghai Jiao Tong University School of Medicine, Xijing Hospital, and Shanghai Songjiang District Central Hospital to host the MICCAI 2019 International Challenge on Gastrointestinal Pathology Image Detection and Segmentation. In addition, Sensetime will collaborate with Yinuo Intelligent Technology and Zhejiang Cancer Hospital to organize the MICCAI 2019 International Challenge on Automatic Structure Delineation for Radiotherapy Planning. These initiatives aim to address clinical needs in the fields of radiotherapy and pathology, open up long-accumulated data resources and expert knowledge to the research community, promote the establishment of relevant evaluation standards, and jointly advance the development of AI in healthcare.

Through in-depth exploration of AI technology and clinical medicine, Sensetime will continue to deliver original technologies, academic achievements, and data resources to the medical field. Leveraging this challenge as a platform, the company aims to further stimulate synergy among industry, academia, and research, promote exchanges between industrial and academic communities, help establish industry norms and standards, and facilitate the clinical implementation of AI in healthcare.

MICCAI (Medical Image Computing and Computer Assisted Intervention) is an internationally recognized premier academic conference with the highest prestige and influence in the fields of medical image computing and computer-assisted intervention. Since 2012, MICCAI has annually hosted Grand Challenges targeting various domains of medical image analysis on a global scale. Leveraging its strong academic leadership, the conference has garnered active responses and participation from top-tier universities and renowned medical technology institutions both domestically and internationally.

Last year, the newly established SenseTime Smart Healthcare team swept the world championships in four major competitions at MICCAI 2018: “Left Ventricle Quantification,” “Left Atrium Segmentation,” “Stroke Lesion Segmentation,” and “Intervertebral Disc Segmentation and Localization.” This achievement demonstrated its research prowess in AI-driven healthcare and provided robust technical support for the subsequent deployment of application scenarios. Bolstered by its proprietary technological capabilities, SenseTime has since focused more on clinical implementation, striving to empower diverse clinical specialties with AI solutions.

Taking radiotherapy and pathology as examples: Sensetime has partnered with MedNo Intelligence, a company with 14 years of expertise in radiotherapy. Building upon MedNo’s radiotherapy planning system, Sensetime leveraged its leading AI capabilities to develop a series of algorithms for organ and target volume delineation. This has effectively improved the efficiency and quality of radiotherapy planning, comprehensively and intelligently enhancing the productivity of limited radiotherapy specialists, thereby benefiting more patients at primary care levels. In the field of pathology, Sensetime collaborated with Hengdao Pathology, China’s first chain-style independent third-party pathology diagnostic center, and jointly worked with multiple hospitals including Shanghai Ruijin Hospital, Songjiang Hospital, and Xi’an Xijing Hospital. They developed algorithms capable of precisely analyzing histopathological features and detecting cellular-level abnormalities. Based on this, they created a comprehensive digestive tract biopsy screening system that accurately and efficiently improves physicians’ workflow, facilitating the practical implementation of AI algorithms in pathology departments.

Through continuous practical implementation, SenseTime has collaborated with hospitals and enterprises to accumulate a large volume of de-identified data (preventing direct or indirect identification of patients' personal information) and, together with experts, has devoted substantial time and effort to high-quality data annotation. This year, SenseTime and its partners will release these high-quality datasets for two MICCAI international challenges in radiotherapy and pathology. In addition to providing datasets larger than those previously available, SenseTime will also offer a long-term open-access evaluation server.

The large-scale standard datasets for automated radiotherapy contouring, signet-ring cell datasets, and colonoscopic tissue section datasets, along with the online testing servers provided by the two competitions, will serve as key benchmark metrics for future algorithms in automated radiotherapy contouring, signet-ring cell detection, and colonoscopic tissue examination, respectively. Artificial intelligence algorithms developed thereafter can be evaluated on these testing servers, undoubtedly endowing the competition with more profound and lasting international influence.

As Zhang Shaoting, Vice President and Deputy Director of the Research Institute at SenseTime, stated: “The development of AI in healthcare is not merely a race for academic achievements and technological breakthroughs, but also a competition in application and scenario implementation. AI technology holds value only when it is truly deployed in practice, effectively helps physicians address critical pain points, tangibly improves their work efficiency and accuracy, and benefits a broader patient population.”

International Challenge I: Automated Structure Delineation for Radiotherapy Planning

Currently, radiotherapy is one of the important means of cancer treatment, and radiotherapy planning is a crucial part of radiation therapy. In order to ensure that the tumor area receives sufficient radiation dose while preventing damage to normal cells in organs at risk due to excessive radiation exposure, it is particularly important to precisely determine the distribution of radiation during the planning phase.

However, in the actual process of formulating radiotherapy plans for patients, physicians are required to delineate organs at risk (OARs) and target volumes across dozens or even hundreds of CT images. This procedure is extremely tedious and labor-intensive, consuming a significant amount of time for radiation oncologists, thereby limiting their work efficiency. In contrast, automatic segmentation algorithms for OARs and target volumes can substantially reduce radiotherapy planning time, enhance physician efficiency, and lower the overall cost of radiotherapy.

However, the industry generally lacks standardized, high-quality training data. The MICCAI 2019 International Challenge on Automatic Structure Segmentation for Radiotherapy Planning, organized this time, will release a large volume of expert-annotated CT data for use in four challenge tasks: “Head and Neck Organs-at-Risk (OAR) Segmentation,” “Thoracic Organs-at-Risk (OAR) Segmentation,” “Radiotherapy Target Volume (GTV) Segmentation for Nasopharyngeal Carcinoma,” and “Radiotherapy Target Volume (GTV) Segmentation for Lung Cancer.”

These CT datasets encompass data for two types of cancer: nasopharyngeal carcinoma and lung cancer. The scale of this dataset surpasses that of all previously publicly available datasets, comprising CT scans from 50 patients with nasopharyngeal carcinoma along with corresponding annotations for 1,100 organs at risk (OARs) and gross tumor volumes (GTVs) for radiotherapy targeting, as well as CT scans from 50 patients with lung cancer along with corresponding annotations for 300 OARs and GTVs. The release of these datasets will significantly advance research on the delineation of organs at risk and target volumes in CT imaging.

International Challenge II: Detection and Segmentation of Gastrointestinal Pathology Images

Pathological examination is the “gold standard” for disease diagnosis. In recent years, with the development of digital pathology, physicians have been able to perform diagnostic analysis on whole-slide images (WSI) remotely, which can significantly improve healthcare standards in remote areas. However, manual analysis of WSI is highly time-consuming for pathologists; some WSI files reach dimensions as large as 100,000 × 100,000 pixels. In regions with an insufficient number of pathologists, the practical implementation of remote WSI examination faces substantial challenges. The development of medical artificial intelligence models to automatically detect, segment, and classify abnormal regions and cells in various pathological images will largely resolve this issue.

The MICCAI 2019 International Challenge on Detection and Segmentation of Gastrointestinal Pathology Images will feature two tasks: “Signet-Ring Cell Detection” and “Colonoscopic Tissue Segmentation and Screening,” aimed at establishing algorithms and evaluation systems for cell detection in digestive system pathology images and biopsy tissue segmentation screening.

This is the world’s first challenge focused on signet-ring cell detection and colonoscopic tissue screening, as well as the first publicly available digestive system pathology image dataset. The dataset released for this competition features extensive expert annotations. The signet-ring cell detection dataset comprises 450 pathological image regions from 90 patients, with 15,000 cell annotations. The colonoscopic tissue segmentation and screening dataset includes 750 pathological image regions from 450 patients, along with corresponding contour annotations of malignant areas. These datasets will significantly advance research in automated pathological object detection and lesion segmentation, further enhancing the application value of artificial intelligence in digital pathology, including clinical decision support and optimization of pathological diagnosis.

In the two major challenges, the algorithmic results submitted by each participating team will be evaluated by top Asian artificial intelligence research institutions, clinical oncology treatment centers, and pathological diagnosis centers. SenseTime has also committed to keeping the test servers provided for this challenge open and supported in the future, allowing researchers to continuously submit and test new methods on these servers.

Within a year, SenseTime has transformed its role from a participant to an organizer of the MICCAI Challenge, a shift driven by its in-depth exploration of AI applications in clinical medical practice. Leveraging big data from real-world implementations and forging strong partnerships with industry institutions, SenseTime fully utilizes its advantages in integrating industry, academia, and research. It is committed to addressing the practical needs of clinicians across the full spectrum of diagnosis, treatment, and recovery workflows, while also sharing its expertise and resources with the broader community to foster overall industry advancement.

Meanwhile, SenseTime aims to deepen collaboration and research with more medical institutions and experts worldwide, promote the establishment of standardized industry norms, jointly advance the development of AI in healthcare, and facilitate the decentralization of high-quality medical resources.

It is reported that the training datasets for the MICCAI 2019 International Challenge on Detection and Segmentation of Gastrointestinal Pathology Images and the MICCAI 2019 International Challenge on Automatic Structure Delineation for Radiotherapy Planning will be released on June 14 and June 15, respectively. The submission deadlines for participating teams are September 23 and September 25, respectively, with the final results to be officially announced on October 1.

Participants can visit the official website to check event details and register for the competition.

MICCAI 2019 International Challenge on Detection and Segmentation of Gastrointestinal Pathology Images

Official Website:https://digestpath2019.grand-challenge.org/Home/

MICCAI 2019 International Challenge on Automatic Structure Segmentation for Radiotherapy Planning Official Website:

https://structseg2019.grand-challenge.org/Home/