.png){kind=icon}
.png){kind=icon}
Return
Digital Biopsy and Network Analysis of Dynamic [68Ga]Ga-FAPI-46 Data in Patients with Malignant and Benign Pancreatic Lesions
.png){kind=icon}
PRE
AI
M
G
A
H
J
J
J
M
M
U
M
M
K
J
U
M
DOI：10.2967/jnumed.125.270185.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The pathologies pancreatic ductal adenocarcinomas, inflammatory lesions of the pancreas, postpancreatectomy reactive tissue, and recurrent pancreatic ductal adenocarcinomas all express fibroblast activation protein and are hardly distinguishable by static PET using [Ga-68]Ga-labeled fibroblast activation protein inhibitors (FAPIs) combined with CT. Dynamic imaging allows full [Ga-68]-Ga-FAPI kinetic profile analysis, highlighting differences among these pathologies. Here, we applied a voxel-level digital biopsy approach combined with network analysis and clustering to characterize healthy, nonmalignant pathologic, and malignant pathologic kinetic signatures. Methods: This monocentric, retrospective study included 47 patients (>18 y) with morphologically unclear pancreatic lesions on CT or MRI and supplemental [Ga-68]Ga-FAPI-46 PET/CT in a primary (31 patients) or recurrent (16 patients) setting. Lesions were classified according to biopsy results (primary cases) or CT appearance and clinical course (recurrent cases). Digital biopsy samples (300 voxels) of pancreatic lesions and control organs (muscle, fat, kidneys, liver, and blood) were taken and then masked and imported into an open source visual analytics application. Voxel networks were created with multiple digital biopsy samples from a single scan or digital biopsy samples combined from multiple scans, with a minimum Pearson correlation value of 0.7. A k-nearest-neighbor edge reduction was applied before Markov clustering. Datasets were then unmasked for interpretation. Static PET parameters (SUVmax and SUVmean) and time to peak of pancreatic lesions and control tissues were extracted from isotropic volumes and analyzed by a t test (threshold for significance, P = 0.05). Results: This work created 47 individual networks and 2 combined networks. Within individual networks, voxels tended to arrange and cluster within the sampled volume of interest (VOI; left and right kidneys strongly coclustered). Networks typically arranged into healthy controls, elimination organs, and pathologic (malignant and nonmalignant) regions. Pathologies tended to cluster with high purity (>95% from the same VOI), with multiple clusters per VOI, indicating intralesional heterogeneity. Our analysis approach could differentiate between malignant and nonmalignant pathologies in the primary and recurrence settings. This differentiation was driven by slower FAPI clearance within malignant voxels. Conclusion: The kinetics of [Ga-68]Ga-FAPI-46 across the different tissues, coupled with this sampling and analysis approach, allowed the separation and identification of healthy, nonmalignant pathologic, and malignant pathologic clusters and kinetic features that may facilitate diagnosis and warrant further investigation.
Keywords:
Ga-[68]Ga-FAPI
pancreatic ductal adenocarcinoma
clustering
dynamic PET
Journal
IF:
9.1
Papers:
6.2K
Citations:
3.0W
Organization
Cited Papers