Introduction
I present a framework for the automatic detection of sagittal cervical spine landmark points, a key objective measure for clinicians managing patients with various symptoms. Drawing inspiration from UNet, I introduce PoseNet, an encoder-decoder Convolutional Neural Network (CNN) designed to address the lack of research in this area. To enhance the accuracy of landmark point localization, I critically examine the limitations of commonly used regression loss functions like L1 and L2 losses. In response, I propose a novel loss function tailored to improve performance in challenging scenarios, such as extreme neck poses, variable brightness, illumination conditions, and X-Ray noise. The effectiveness of the proposed framework and loss function is validated using a dataset of X-Ray images, demonstrating precise detection of sagittal cervical spine landmark points even in challenging image conditions.
This research is the part of PostureRay.
