Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AutoPET Challenge III: Testing the Robustness of Generalized Dice Focal Loss trained 3D Residual UNet for FDG and PSMA Lesion Segmentation from Whole-Body PET/CT Images

Shadab Ahamed

TL;DR

A 3D Residual UNet model is utilized and the Generalized Dice Focal Loss function is employed to train the model on the AutoPET Challenge 2024 dataset, using an average ensembling technique using the models from the five folds.

Abstract

Automated segmentation of cancerous lesions in PET/CT scans is a crucial first step in quantitative image analysis. However, training deep learning models for segmentation with high accuracy is particularly challenging due to the variations in lesion size, shape, and radiotracer uptake. These lesions can appear in different parts of the body, often near healthy organs that also exhibit considerable uptake, making the task even more complex. As a result, creating an effective segmentation model for routine PET/CT image analysis is challenging. In this study, we utilized a 3D Residual UNet model and employed the Generalized Dice Focal Loss function to train the model on the AutoPET Challenge 2024 dataset. We conducted a 5-fold cross-validation and used an average ensembling technique using the models from the five folds. In the preliminary test phase for Task-1, the average ensemble achieved a mean Dice Similarity Coefficient (DSC) of 0.6687, mean false negative volume (FNV) of 10.9522 ml and mean false positive volume (FPV) 2.9684 ml. More details about the algorithm can be found on our GitHub repository: https://github.com/ahxmeds/autosegnet2024.git. The training code has been shared via the repository: https://github.com/ahxmeds/autopet2024.git.

AutoPET Challenge III: Testing the Robustness of Generalized Dice Focal Loss trained 3D Residual UNet for FDG and PSMA Lesion Segmentation from Whole-Body PET/CT Images

TL;DR

A 3D Residual UNet model is utilized and the Generalized Dice Focal Loss function is employed to train the model on the AutoPET Challenge 2024 dataset, using an average ensembling technique using the models from the five folds.

Abstract

Automated segmentation of cancerous lesions in PET/CT scans is a crucial first step in quantitative image analysis. However, training deep learning models for segmentation with high accuracy is particularly challenging due to the variations in lesion size, shape, and radiotracer uptake. These lesions can appear in different parts of the body, often near healthy organs that also exhibit considerable uptake, making the task even more complex. As a result, creating an effective segmentation model for routine PET/CT image analysis is challenging. In this study, we utilized a 3D Residual UNet model and employed the Generalized Dice Focal Loss function to train the model on the AutoPET Challenge 2024 dataset. We conducted a 5-fold cross-validation and used an average ensembling technique using the models from the five folds. In the preliminary test phase for Task-1, the average ensemble achieved a mean Dice Similarity Coefficient (DSC) of 0.6687, mean false negative volume (FNV) of 10.9522 ml and mean false positive volume (FPV) 2.9684 ml. More details about the algorithm can be found on our GitHub repository: https://github.com/ahxmeds/autosegnet2024.git. The training code has been shared via the repository: https://github.com/ahxmeds/autopet2024.git.
Paper Structure (13 sections, 5 equations, 5 figures, 2 tables)

This paper contains 13 sections, 5 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials and methods
  3. Data and data split
  4. Preprocessing and data augmentation
  5. Network
  6. Loss function, optimizer, and scheduler
  7. Inference and postprocessing
  8. Evaluation metrics
  9. Results and discussion
  10. Five-fold cross validation results and the preliminary test phase
  11. Performance across FDG vs. PSMA cases
  12. Conclusion
  13. Acknowledgment

Figures (5)

  • Figure 1: Distribution of DSC (left), FNV (middle) and FPV (right) from FDG (blue) and PSMA (orange) cases across each of the validation folds. The legend shows the mean values of the metrics over the two radiotracers. On an average, the networks performance on the FDG cases were superior than those on the PSMA cases across all folds, especially on the DSC and FPV metrics.
  • Figure 2: Ground truth lesion-level lesion measures such as lesion MTV (left), SUV$_\text{mean}$ (middle), and SUV$_{\text{max}}$ (right) across FDG (blue) and PSMA (orange) cases in the training set.
  • Figure 3: Ground truth patient-level lesion measures such as TMTV (left), TLG (middle), and number of lesions (right) across FDG (blue) and PSMA (orange) cases in the training set.
  • Figure 4: Some example cases showing the comparison between the ground truth and predicted lesion segmentation masks for four cases of lymphoma (left) and four cases of lung cancer (right).
  • Figure 5: Some example cases showing the comparison between the ground truth and predicted lesion segmentation masks for four cases of melanoma (left) and four cases of prostate cancer (right).