# Combining Image- and Geometric-based Deep Learning for Shape Regression: A Comparison to Pixel-level Methods for Segmentation in Chest X-Ray

arXiv: https://arxiv.org/abs/2401.07542 · HTML: https://mlsanity.com/paper/2401.07542/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 7 of its 12 references that are in the map.
- First posted on arXiv 15 Jan 2024.

## Builds on (most cited first, 7 of 7)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation](https://mlsanity.com/paper/1612.00593/) (2016)
- [Searching for MobileNetV3](https://mlsanity.com/paper/1905.02244/) (2019)
- [Point Transformer](https://mlsanity.com/paper/2011.00931/) (2020)
- [PolyTransform: Deep Polygon Transformer for Instance Segmentation](https://mlsanity.com/paper/1912.02801/) (2019)
- [Annotating Object Instances with a Polygon-RNN](https://mlsanity.com/paper/1704.05548/) (2017)
- [Human pose estimation via Convolutional Part Heatmap Regression](https://mlsanity.com/paper/1609.01743/) (2016)

## Built on by (most cited first, 1 of 1)

- [DenseSeg: Joint Learning for Semantic Segmentation and Landmark Detection Using Dense Image-to-Shape Representation](https://mlsanity.com/paper/2405.19746/) (2024)

## Abstract (arXiv)

When solving a segmentation task, shaped-base methods can be beneficial compared to pixelwise classification due to geometric understanding of the target object as shape, preventing the generation of anatomical implausible predictions in particular for corrupted data. In this work, we propose a novel hybrid method that combines a lightweight CNN backbone with a geometric neural network (Point Transformer) for shape regression. Using the same CNN encoder, the Point Transformer reaches segmentation quality on per with current state-of-the-art convolutional decoders (4±1.9 vs 3.9±2.9 error in mm and 85±13 vs 88±10 Dice), but crucially, is more stable w.r.t image distortion, starting to outperform them at a corruption level of 30%. Furthermore, we include the nnU-Net as an upper baseline, which has 3.7× more trainable parameters than our proposed method.