Computer Vision / Video Analytics

Pearl Raises $11 Million to Develop an AI-Based Dental Analysis Technology

Santa Monica, California-based startup Pearl, a new healthcare company focused on the dental industry, has just raised $11 million in series A funding to create a holistic oral health platform.

“Pearl will have an immediate positive impact on the dental category,” Ophir Tanz, the company’s CEO told VentureBeat.  “It will streamline tedious, repetitive tasks, enhance profitability across dentistry, and, most importantly, it will improve the standard of care by validating diagnoses, removing large elements of uncertainty from the dental equation.”

The company says the dental business can help reduce insurance fraud, validate the performance of dentists, and help automate workflows inside a dental office.

On the backend, the company is using NVIDIA V100 GPUs and GeForce RTX 2080 TI GPUs with the cuDNN-accelerated TensorFlow and PyTorch deep learning frameworks to train their neural networks. Cambron Carter, the company’s Co-Founder and Chief Technology Officer says the training data is on the order of millions of x-ray images.

inference is done on NVIDIA Tesla GPUs on the Amazon Web Services cloud, Cambron says.

At launch, the company offers four tools to address different dental issues.

“Second Opinion”: The system uses hundreds of dentist annotated images to help identify dozens of common pathologies in dental x-rays. “Practice Intelligence”: can help manage practice needs such as marketing, hiring, ROI trends, as well as on-demand performance reports that highlight trends in patient pools. “Smart Margin” and “Scan Clarity”: Helps score dental scans. The system can also mark scans for contact points and defects. High-margin scans are sent to the next step in the process, while low-margin scans are flagged for human review.

Dental Scan Labeled by Pearl

Cambron says they are currently exploring the use of GANs for refining training data quality as well as designing architectures to segment 3D intraoral meshes.

Read more>

Discuss (0)