Machine Learning for Medical Robotics:

• Implemented a Mask R-CNN instance segmentation model using ONNX Runtime in C++ to detect and identify fiducial markers in the camera feed of the ARTAS robot, validating model outputs against expected 3D spatial locations and marker IDs to ensure accurate registration.

• Implemented a Faster R-CNN object detection model using ONNX Runtime in C++ to perform real-time detection of existing scalp extraction sites from the camera feed of the ARTAS robot, dynamically blocking those regions to prevent repeat harvesting

• Leveraged Azure ML Studio to train instance segmentation models for real-time detection and shape identification of the extraction needle.

• Trained convolutional neural networks (CNNs) using TensorFlow to classify post-extraction images and assess hair follicle extraction success.

• Created a custom image labeling tool to annotate and extract thousands of images for training.

User Interface Development:

• Modernized and streamlined the ARTAS UI using WPF and Caliburn.Micro, making workflows more intuitive and reducing opportunities for user error.

• Collaborated with a graphic designer to improve the UI’s visual design and overall user experience.

Graphics & Visualization:

• Upgraded the robot’s graphics pipeline to modern OpenGL (v3.3), enabling support for rendering 3D models, dynamic overlays, and visual treatment previews.

• Developed features such as real-time 3D rendering of hair follicles over live camera feeds to assist in treatment planning.

Touchscreen Interaction & Connectivity:

• Designed and implemented a touch-based algorithm for users to draw treatment areas directly on the UI, improving workflow efficiency.

• Enabled Wi-Fi connectivity using Tik4Net, reducing dependency on wired connections for setup and support.