Sine-Wave-Normalization-for-Deep-Learning-Based-Tumor-Segmentation-in-CT-PET

Sine-Wave-Transformations-for-Deep-Learning-Based-Tumor-Segmentation-in-CT-PET-Imaging

This repository contains the deep learning model for automated tumor lesion segmentation in CT/PET scans, designed for the autoPET III Challenge. It introduces a novel SineNormalBlock, leveraging sine wave transformations to enhance PET data processing for improved segmentation accuracy.

The code is publicly available at: Sine-Wave Normalization for Deep Learning Based Tumor Segmentation in CT-PET

Below is the shared nnUNet trained model folder structure.

├── nnUNet_results
│   └── Dataset101_autopet
│       └── nnUNetTrainerUmambaSinNorm__nnUNetResEncUNetMPlans__3d_fullres_bs8
│           ├── dataset_fingerprint.json
│           ├── dataset.json
│           ├── fold_all
│           │   ├── checkpoint_best.pth
│           │   └── debug.json
│           └── plans.json
└── README.md