Optimizing Segmentation of Medical Images

Paper Title:

Enhancing Medical Image Segmentation: Optimizing Cross-Entropy Weights and Post-Processing with Autoencoders

Published on:

21 August 2023

Primary Category:

Image and Video Processing

Paper Authors:

Pranav Singh,

Luoyao Chen,

Mei Chen,

Jinqian Pan,

Raviteja Chukkapalli,

Shravan Chaudhari,

Jacopo Cirrone

Bullets

Key Details

•

Presents deep learning approach for medical image segmentation

•

Focuses on segmenting images of the rare disease dermatomyositis

•

Uses U-Net and U-Net++ architectures with optimized loss weights

•

Incorporates autoencoder post-processing step

•

Outperforms prior state-of-the-art for dermatomyositis segmentation by over 12%

•

Benchmarks approach on two additional medical imaging datasets

Explore the topics in this paper

deep learning

dermatomyositis

medical imaging

rare disease

segmentation

AI generated summary

Optimizing Segmentation of Medical Images

This paper presents a deep learning approach to segment medical images, focusing on a rare autoimmune disease called dermatomyositis. The method uses U-Net and U-Net++ architectures and outperforms prior state-of-the-art techniques by over 12% in segmenting images of dermatomyositis histopathology slides. The authors optimize the loss function weights and incorporate autoencoder post-processing. They benchmark the approach on two additional challenging medical imaging datasets. Overall, the work demonstrates significant advances in medical image segmentation, especially for rare diseases where data is limited.