Automated Pipeline for Longitudinal Lesion Tracking in SPECT Imaging using Morphology and Texture Aware Cost Function

Longitudinal Lesion Tracking (SPECT/CT)

Robust longitudinal lesion tracking for nuclear medicine imaging (SPECT/CT):

• Rigid registration T2 → T1 (SimpleITK) with optional histogram matching and foreground body masks
• Geometry-safe mask alignment (header fix vs. NN resample) to prevent silent mis-registrations
• Feature construction: centroids (mm), volumes (mm³), simple heterogeneity (PyRadiomics first-order + GLCM)
• Partial Hungarian matching with spatial gating + cosine feature similarity
• Visualization: joint-window coronal MIPs, links colored by match strength, explicit missed/new lesions
• QC: blended MIPs with Pearson correlation and MAD metrics
• CLI & notebook-friendly: clean logging, deterministic seeds, clear outputs

Use case: Longitudinal SPECT/CT (e.g., RPT monitoring) where consistent lesion correspondence is needed to form lesion-level TACs, estimate TIA, and enable dose–response modeling.

---

Citation

If you use this code, please cite:

Yousefirizi F, Kurkowska S, Benard F, Rahmim A, Uribe C.
Automated Pipeline for Longitudinal Lesion Tracking in SPECT Imaging using Morphology and Texture Aware Cost Function. 2025.

Install necessary libraries

SimpleITK>=2.3.1 pyradiomics>=3.1.0 numpy>=1.26 pandas>=2.0 scipy>=1.11 scikit-learn>=1.3 matplotlib>=3.8

Inputs

The pipeline requires two timepoints (T1 and T2) for the same patient:

• SPECT/CT images

  • spect.nii.gz (or another name specified in spect-name)
  • Must be in NIfTI format, with consistent orientation/spacing

• Lesion masks

  • Binary mask files (.nii or .nii.gz) located in the lesions/ subfolder for each timepoint
  • Each mask corresponds to one lesion (largest connected component is extracted)
  • Lesion ID is parsed from the filename (e.g., lesion_01.nii.gz → ID=01)

Example folder structure:

patient_folder/
  ├─ 2024-06-29/               # T1
  │   ├─ spect/spect.nii.gz
  │   └─ lesions/lesion_01.nii.gz, lesion_02.nii.gz, ...
  └─ 2024-10-05/               # T2
      ├─ spect/spect.nii.gz
      └─ lesions/lesion_01.nii.gz, lesion_02.nii.gz, ...