Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Local Gamma Augmentation for Ischemic Stroke Lesion Segmentation on MRI

Jon Middleton, Marko Bauer, Kaining Sheng, Jacob Johansen, Mathias Perslev, Silvia Ingala, Mads Nielsen, Akshay Pai

TL;DR

The paper tackles data scarcity in ischemic stroke lesion segmentation on multi-modal MRI by introducing local gamma augmentation, a targeted intensity transformation applied only to pathologic regions identified by segmentation maps. The method defines a local gamma transform and blends augmented pathology back into normal tissue using a mask, with gamma parameters drawn from a fixed mixture distribution. Across three private MRI datasets, the approach yields consistent gains in image-level sensitivity (e.g., increases from 0.566 to 0.736 on the WUS set) with minimal or acceptable changes to specificity, demonstrating improved detection of ischemic lesions under limited voxel-level annotations. The contribution provides a simple, easily integrable augmentation technique that mitigates intensity bias and enhances data efficiency for pathology segmentation tasks in MRI, with potential applicability to other diseases characterized by region-specific intensity differences.

Abstract

The identification and localisation of pathological tissues in medical images continues to command much attention among deep learning practitioners. When trained on abundant datasets, deep neural networks can match or exceed human performance. However, the scarcity of annotated data complicates the training of these models. Data augmentation techniques can compensate for a lack of training samples. However, many commonly used augmentation methods can fail to provide meaningful samples during model fitting. We present local gamma augmentation, a technique for introducing new instances of intensities in pathological tissues. We leverage local gamma augmentation to compensate for a bias in intensities corresponding to ischemic stroke lesions in human brain MRIs. On three datasets, we show how local gamma augmentation can improve the image-level sensitivity of a deep neural network tasked with ischemic lesion segmentation on magnetic resonance images.

Local Gamma Augmentation for Ischemic Stroke Lesion Segmentation on MRI

TL;DR

The paper tackles data scarcity in ischemic stroke lesion segmentation on multi-modal MRI by introducing local gamma augmentation, a targeted intensity transformation applied only to pathologic regions identified by segmentation maps. The method defines a local gamma transform and blends augmented pathology back into normal tissue using a mask, with gamma parameters drawn from a fixed mixture distribution. Across three private MRI datasets, the approach yields consistent gains in image-level sensitivity (e.g., increases from 0.566 to 0.736 on the WUS set) with minimal or acceptable changes to specificity, demonstrating improved detection of ischemic lesions under limited voxel-level annotations. The contribution provides a simple, easily integrable augmentation technique that mitigates intensity bias and enhances data efficiency for pathology segmentation tasks in MRI, with potential applicability to other diseases characterized by region-specific intensity differences.

Abstract

The identification and localisation of pathological tissues in medical images continues to command much attention among deep learning practitioners. When trained on abundant datasets, deep neural networks can match or exceed human performance. However, the scarcity of annotated data complicates the training of these models. Data augmentation techniques can compensate for a lack of training samples. However, many commonly used augmentation methods can fail to provide meaningful samples during model fitting. We present local gamma augmentation, a technique for introducing new instances of intensities in pathological tissues. We leverage local gamma augmentation to compensate for a bias in intensities corresponding to ischemic stroke lesions in human brain MRIs. On three datasets, we show how local gamma augmentation can improve the image-level sensitivity of a deep neural network tasked with ischemic lesion segmentation on magnetic resonance images.
Paper Structure (14 sections, 4 equations, 2 figures, 2 tables)

This paper contains 14 sections, 4 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Gamma transformations
  5. Local gamma augmentations
  6. Data
  7. Training Data
  8. Test Data
  9. Experiments and Results
  10. Experimental Setup
  11. Evaluation
  12. Results
  13. Discussion
  14. Conclusion

Figures (2)

  • Figure 1: Depiction of local gamma augmentation. An image $I(x)$ and its gamma transformation $I_\gamma(x)$ are multiplied pointwise with a segmentation map and combined to produce an image $I_{M, \gamma}(x)$ featuring an augmentation of a pathology.
  • Figure 2: Gamma transformations of a diffusion weighted image from the ISLES-2022 dataset. The center column shows the original image. The top row contains globally gamma compressed images to the left and globally gamma expanded images to the right. The bottom row consists of the same gamma transformations restricted to the location of an acute ischemic stroke.