HEALTHCARE

Medical Video Processing AI

Realtime video processing AI for identifying the crucial moments and Clinical View of Safety in a surgical setting. Artificial Intelligence is enhanced and deployed on the computer on wheels (CoW) hardware to assist surgeons during the operation.

Results | Background | What we did

Results

We worked with the client to identify the best approach and tech stack. The state-of-the-art deep learning model and pipeline were explored to craft a future-proof solution.

Leveraged Specialized Expertise

Utilized the fundamental knowledge of computer vision and machine learning pipeline architecture to apply it to a surgical use case.

Modular Integration

The streaming pipeline supports variants of inputs and AI models to provide future-proof flexibility.

Performance & Accuracy Focused

Achieved low-latency with high accuracy, one of the biggest challenges in practically applying AI.

Background

Curantis, Dallas, TX

Enabling real-time video processing AI for assisting laparoscopic cholecystectomy procedure on human patients.

Team

Project/product mgmt.
Architecture
Development

Build Type

Greenfield

Tech Stack

NVidia DeepStream + Triton Inference Server
Jetson (Linux on ARM)
NVidia Clara AGX

What We Did

Exploration & Planning

We worked with the client to:

Identify the suitable technology stack for low-latency and modular architecture to support multi-input, multi-model capability.
Plan for the development of the video streaming pipeline which will happen concurrently with the improvement of the deep learning AI.

Design & Development

Design and develop the pipeline for the continuous real-time execution AI on the provided CoW hardware. We then measured the baseline performance.

Plug & Play

Merge the pipeline with the deep learning models hosted on NVidia’s Triton Inference Server. The tuning and testing are supported to ensure that we satisfy the real-world usage requirements.