VoxelCloud, a startup based out of Los Angeles and with a presence in Suzhou and Shanghai, China, has developed a suite of artificial intelligence and cloud computing technologies to assist doctors in interpreting medical images. The technology provides fully automated medical imaging analysis, and can be used with various imaging techniques, such as computed tomography or digital color imaging of the retina. At present, the system has been developed for use in diagnosing lung cancer, retinal diseases, and coronary heart disease.
The platform is designed to complement and assist a clinician in their decision-making process. The company is currently in the process of expanding the scope of its services, using some recently acquired funding. VoxelCloud has worked with a variety of healthcare institutions in China and the US to trial the technology.
Medgadget had the opportunity to ask Xiaowei Ding, CEO of VoxelCloud, some questions about the technology.
Conn Hastings, Medgadget: Please introduce yourself and tell us a little about how you got into this area.
Xiaowei Ding, VoxelCloud: I’m the co-founder and CEO of VoxelCloud. I started VoxelCloud in 2015 in Los Angeles. Actually, VoxelCloud is an extension of my research program, which I started with my PhD research.
After I finished my bachelor’s degree in Electrical Engineering from Shanghai Jiao Tong University and a PhD in Computer Science from UCLA in 2015, I decided to apply my background in computer science to the medical industry. In fact, during my PhD research in the computer science lab at UCLA, I had explored the possibilities that medicine could benefit from the development of computer sciences. Later on, I worked at the Cedars-Sinai Medical Center as a research assistant in the Biomedical Imaging Research Institute and the Artificial Intelligence in Medicine Program.
During my research, I found that before 2013, AI in medical research programs has been limited by the amount of available data, and there were not enough good machine-learning algorithms to apply those clinical data. I want to solve this problem.
So, after one year of investigation, I co-founded VoxelCloud, a startup that aims at providing medical image analysis and diagnosis assistance for clinical practices. With artificial intelligence and cloud computing technologies, we can teach machines to understand medical data and boost a doctor’s efficiency and diagnostic confidence.
Medgadget: Put simply, how does the technology work? What types of imaging does the system work with?
Xiaowei Ding: Intuitively, we present our algorithms with a vast amount of medical data and the corresponding diagnosis. During the training process, we let the model learn the rules to arrive at such insights from the original data.
In clinical practice, this information will assist doctors to make informed clinical decisions more efficiently. So far, we have applied this methodology to x-ray computed tomography and fundus photography images, and there is no theoretical limitation to where our technology can be applied.
Medgadget: What problem does the system solve? Is it currently difficult to interpret these images accurately?
Xiaowei Ding: Interpreting medical images is a highly specialized profession that requires the practitioners to undergo many years of extensive training. The supply of highly trained doctors is being outpaced by the ever-growing demand for image interpretation. When under time pressures, providing an accurate diagnosis may become an issue for many human readers.
Our technology aims to help doctors interpret medical images more efficiently and make more informed clinical decisions through our data-driven, algorithmic insights. In turn, we are also making accurate diagnosis more accessible to a greater audience.
Medgadget: Please tell us a little about your plans to expand the scope of the services VoxelCloud offers.
Xiaowei Ding: Our service will include at least three aspects. First, we will provide a service directly to healthcare providers (such as hospitals, medical centers, etc.) through clinical workflow integrated cloud-computing solutions. Second, we will partner with existing hardware and software vendors to provide an AI service. Third, we will build a platform to enable third party medical developers to develop their own applications much more easily through our medical knowledge graph API. With this API, developers could work more efficiently because we have developed the basic algorithms and collected a massive amount of anonymized training data from scratch. Our medical image knowledge graph will be offered though this API. Further extension of our project will leverage heterogeneous data sources in synergy with the existing imaging data.
Medgadget: How and where is the system used at the moment? Have clinicians found it helpful?
Xiaowei Ding: We are working side by side with the leading medical institutions in China to actively develop and evaluate our lung cancer screening and eye disease diagnosis products. We have received very positive feedback from the physicians we are working with.
The clinicians have found that our products can boost workflow efficiency and have the potential to bring more accurate, imaging-based disease screening and diagnosis through our AI and cloud-computing platform.
Medgadget: What are the limits of AI for these types of applications? Do you ever see the system being a replacement for a clinician, or always an assistive technology?
Xiaowei Ding: The performance of an AI system is only as good as the data it has seen. Collecting and utilizing a large amount of high-quality medical data is a big challenge and will be a challenge in the near term, because there are patient privacy concerns and a lack of digitized records in many parts of the world.
This also means, that in many cases, the ability of an AI system in medicine can be limited to a narrow scope. To make a well-informed diagnosis, a doctor will have to consider all contexts of the patient’s situation. Therefore, we don’t foresee AI systems to be a replacement for a clinician, but will be a very reliable aid to clinicians, because they allow clinicians to delegate the repetitive, manual tasks to the machines and focus their precious time on what they do best, which is gathering all necessary information and deciding on the best solution for the patient.
Link: VoxelCloud…
