What they are?
Medical data is being generated at an exponential rate, from medical pictures to connected devices. Traditional diagnostic procedures are rendered obsolete as a result of this, as well as new types of data generated by genome sequencing and biosensors. Qure has a team of dedicated computer scientists, medical practitioners, and bioinformaticians to use deep learning to make healthcare more accessible and inexpensive. They believe that Artificial Intelligence (AI) is the answer to guaranteeing that healthcare practitioners can focus on the cases that matter, while machines diagnose and treat the ones that aren’t as important. Their goal is to use deep learning to identify disease using radiology and pathology imaging, as well as to develop individualized cancer therapy programs using psychopathology imaging and genetic sequences.
Their study is carried out in partnership with several hospital systems, universities, and research organizations.
In 2016, Qure.ai was established. Their goal is to make healthcare more accessible and inexpensive by utilizing artificial intelligence. Deep learning expertise is combined with clinical, scientific, and regulatory understanding in their core team. Radiologists, other doctors, and public health professionals make up their advisory panel. They collaborate with these experts to identify clinically relevant issues and provide practical solutions.
Highlights of their Research
Visualizing what deep neural networks learn in interpretable ai
When working with AI, doctors like to see a region-of-interest or a visual representation of which sections of the image the algorithm depends on the most when making a decision. This allows individuals to perceive the world through the eyes of the algorithm. Area markers or heatmaps like these give clinical users visual indications that can help them decide whether to accept or reject a chest x-ray finding reported by AI.
Deep learning algorithm interpretability is a hot topic in research — and a focus for Qure.ai. There are two types of visualization approaches now available: perturbation-based visualizations and backpropagation-based visualizations. They’ve tried out these techniques and created a blog series about how they work with medical photos, using chest X-rays as an example.
Validation study and open data set
For patients with head trauma or stroke symptoms, a head CT scan is the conventional initial imaging study. They conducted research to assess the accuracy of a set of deep learning algorithms that were trained to detect head CT anomalies that required immediate treatment (against three radiologists). Intraparenchymal (IPH), intraventricular (IVH), subdural (SDH), extradural (EDH), and subarachnoid (SAH) intracranial hemorrhages (ICH) and skull fractures are all detected by the learning algorithms. They also detect mass effect and midline displacement, both of which are utilized as markers of brain injury severity.
Natural language processing for radiology reports
Large volumes of well-labeled training data are required to build accurate deep learning models. The most scalable way to provide classification algorithms with the data they require to achieve high accuracy is to use the accompanying clinical reports as ground truth for radiological scans.
However, instead of being prepared systematically, these reports are usually produced as free-form text. It’s critical to construct sophisticated natural language processing (NLP) algorithms to tap into the expertise included in radiology reports since it allows them to create ground truth labels at scale (hundreds of thousands to millions of scans).
Rule-based NLP systems parse and arrange unstructured text using a set of manually specified rules. When taught on a large annotated dataset, Machine Learning (ML) based NLP systems, on the other hand, build the rules automatically. To read radiology reports, they designed a bespoke dictionary-based NLP system and compared it to machine-learning alternatives.
Prashant Warier, Co-Founder & CEO
Prashant is a specialist in Artificial Intelligence and Deep Learning. In his 19-year career, he has designed and marketed several data science products. He is also a well-known researcher, author, and speaker on data science and machine learning themes. He is dedicated to making healthcare more inexpensive and accessible through the use of deep learning.
|Company size||51-200 employees|