Our Solution

Overview

Unlike market alternatives, OralVision utilizes advanced machine learning to analyze intraoral photos and provide a detailed assessment of possible patient conditions. Furthermore, our innovation is significantly more affordable than conventional devices, making it more accessible to underprivileged communities.

How It Works

Instead of simply illuminating abnormal tissue, OralVision takes multiple photos of a patient’s oral cavity and feeds them into a comprehensive federated neural network that our company developed and trained. Our innovation has the capability to provide highly accurate clinical diagnoses and iteratively refine model parameters based on aggregated HIPAA-compliant patient data, giving us a significant edge over market competitors that haven’t even considered integrating machine learning with intraoral diagnostics yet.

In addition, OralVision utilizes cost-effective hardware, enabling our company to produce it at a relatively lower marginal cost and sell it at a highly competitive price that’s more accessible to underserved communities. The only notable disadvantage that OralVision has in comparison with existing solutions is the fact that the device is quite bulky and not ergonomically optimized. However, we have plans to make the device frame more streamlined to maximize patient comfort in future prototypes.

Federated Learning

Yet, unlike other devices, OralVision’s machine learning algorithm can be trained and updated using federated learning, where physicians can mark real-world responses when they differ from what our model outputs. Modified weights will be sent to a global model which can be used on all client devices, improving the performance of our models.

We will only be transferring the new model weights given by client devices and not any images or sensitive information, maintaining HIPAA compliance while continuously improving the primary function of the device. Additionally, physicians can update their OralVision models quickly and easily over the Internet and will be reminded periodically using the system’s internal clock to update the device’s weights. However, in low-income areas where access to the Internet may not always be feasible, the devices can continue to work with older model architectures without updating.

Ease of Use

OralVision attempts to lower the ceiling for the diagnosis of oral malignancies by streamlining the processing pipeline. Users are not required to have any prior medical experience to utilize the device, as the minimal control system in conjunction with the software-level abstraction ensures a smooth user experience.

Moreover, OralVision was designed with the limitations of its most prominent environment in mind, as it will be shipped with features such as rechargeable batteries and solar power capabilities pre-installed, lowering the barrier to usage and expediting the setup timeline. OralVision enables physicians in low-income areas without the practical nor technical experience in dentistry or stomatology to diagnose their patients effortlessly.

Competitive Protection

In order to maintain our device’s proprietary aspects, we will be privatizing our machine learning model and its weights so that our competitors cannot utilize the data we have. By using federated learning, we can ensure our model consistently stays ahead of competitors by using private data.

Additionally, we will patent the instrument and the processor’s design to prevent others from attempting to create similarly cost-effective devices. We will also be aggressively expanding in our early years to improve our model and incorporate our device to as many places as possible, gaining a competitive edge over the rest of the market.