Training on Polar Image Transformations Improves Biomedical Image Segmentation

Abstract

A key step in medical image-based diagnosis is image segmentation. A common use case for medical image segmentation is the identification of single structures of an elliptical shape. Most organs like the heart and kidneys fall into this category, as well as skin lesions, polyps, and other types of abnormalities. Neural networks have dramatically improved medical image segmentation results, but still require large amounts of training data and long training times to converge. In this paper, we propose a general way to improve neural network segmentation performance and data efficiency on medical imaging segmentation tasks where the goal is to segment a single roughly elliptically distributed object. We propose training a neural network on polar transformations of the original dataset, such that the polar origin for the transformation is the center point of the object. This results in a reduction of dimensionality as well as a separation of segmentation and localization tasks, allowing the network to more easily converge. Additionally, we propose two different approaches to obtaining an optimal polar origin: (1) estimation via a segmentation trained on non-polar images and (2) estimation via a model trained to predict the optimal origin. We evaluate our method on the tasks of liver, polyp, skin lesion, and epicardial adipose tissue segmentation. We show that our method produces state-of-the-art results for lesion, liver, and polyp segmentation and performs better than most common neural network architectures for biomedical image segmentation. Additionally, when used as a pre-processing step, our method generally improves data efficiency across datasets and neural network architectures.

Using the Polar Transform for Efficient Deep Learning-Based Aorta Segmentation in CTA Images

“…We show that this method is comparable to the state-of-theart aorta segmentation methods while being robust to small dataset sizes. In addition, we extend the method presented in [1] and further validate the use of polar transformations in neural networks for medical image segmentation. These modifications can be used to improve performance in a wide variety of medical image processing tasks where multiple elliptical objects need to be segmented, and can be added to existing methods for 2D image semantic segmentation without changing the underlying architecture.…”

“…We then sum all of the weighted images together. As shown in [1], the polar network generally performs best on objects which contain the polar origin, and worse at predicting other objects on the image. Therefore, we assign a larger weight to that component as a proxy for a confidence measure.…”

“…The work presented in this paper is an extension of [1]. We obtain the center points of the objects in the image using a rough segmentation from a U-Net-based network, instead of using a center point predictor as is described in the original paper.…”

“…The method presented in this paper is an extension and improvement over the method presented in [1]. The method uses a neural network to predict the center point of a single object in an input image.…”

“…All models were trained for 60 epochs with checkpointing where the model with the best validation Dice coefficient was selected. We use the Dice loss function as described in [1]. All of the code, as well as the trained networks, can be found at github.com/marinbenc/medical-polar-training…”