Binary segmentation of medical images using implicit spline representations and deep learning

TL;DR

This paper introduces a new deep learning-based method for medical image segmentation that uses implicit spline representations to accurately delineate boundaries, achieving state-of-the-art results on a heart disease dataset.

Contribution

It combines implicit spline representations with deep neural networks for segmentation, introducing novel loss functions and adapting existing architectures for this purpose.

Findings

Spline of bidegree (1,1) with 128x128 coefficients performed best.

Achieved an average Dice score of nearly 92% on the test set.

Method outperforms previous approaches on the congenital heart disease dataset.

Abstract

We propose a novel approach to image segmentation based on combining implicit spline representations with deep convolutional neural networks. This is done by predicting the control points of a bivariate spline function whose zero-set represents the segmentation boundary. We adapt several existing neural network architectures and design novel loss functions that are tailored towards providing implicit spline curve approximations. The method is evaluated on a congenital heart disease computed tomography medical imaging dataset. Experiments are carried out by measuring performance in various standard metrics for different networks and loss functions. We determine that splines of bidegree $(1,1)$ with $128\times128$ coefficient resolution performed optimally for $512\times 512$ resolution CT images. For our best network, we achieve an average volumetric test Dice score of almost 92%, which reaches the state of the art for this congenital heart disease dataset.