An automated pheochromocytoma and paraganglioma lesion segmentation AI-model at whole-body 68Ga- DOTATATE PET/CT

Abstract

Background: Somatostatin receptor (SSR) targeting radiotracer ⁶⁸Ga-DOTATATE is used for Positron Emission Tomography (PET)/Computed Tomography (CT) imaging to assess patients with Pheochromocytoma and paraganglioma (PPGL), rare types of Neuroendocrine tumor (NET) which can metastasize thereby becoming difficult to quantify. The goal of this study is to develop an artificial intelligence (AI) model for automated lesion segmentation on whole-body 3D DOTATATE-PET/CT and to automate the tumor burden calculation. 132 ⁶⁸Ga-DOTATATE PET/CT scans from 38 patients with metastatic and inoperable PPGL, were split into 70, and 62 scans, from 20, and 18 patients for training, and test sets, respectively. The training set was further divided into patient-stratified 5 folds for cross-validation. 3D-full resolution nnUNet configuration was trained with 5-fold cross-validation. The model's detection performance was evaluated at both scan and lesion levels for the PPGL test set and two other clinical cohorts with NET (n = 9) and olfactory neuroblastoma (ONB, n = 5). Additionally, quantitative statistical analysis of PET parameters including SUVmax, total lesion uptake (TLU), and total tumor volume (TTV), was conducted.

Results: The nnUNet AI model achieved an average 5-fold validation dice similarity coefficient of 0.84 at the scan level. The model achieved dice similarity coefficients (DSC) of 0.88, 0.6, and 0.67 at the scan level, the sensitivity of 86%, 61.13%, and 61.64%, and a positive predictive value of 89%, 74%, and 86.54% at the lesion level for the PPGL test, NET and ONB cohorts, respectively. For PPGL cohorts, smaller lesions with low uptake were missed by the AI model (p < 0.001). Anatomical region-based failure analysis showed most of the false negative and false positive lesions within the liver for all the cohorts, mainly due to the high physiologic liver background activity and image noise on ⁶⁸Ga- DOTATATE PET scans.

Conclusions: The developed deep learning-based AI model showed reliable performance for automated segmentation of metastatic PPGL lesions on whole-body ⁶⁸Ga-DOTATATE-PET/CT images, which may be beneficial for tumor burden estimation for objective evaluation during therapy follow-up. https://www.

Clinicaltrials: gov/study/NCT03206060 , https://www.

Clinicaltrials: gov/study/NCT04086485 , https://www.

Clinicaltrials: gov/study/NCT05012098 .

Keywords: Artificial intelligence; DOTATATE PET/CT; Deep learning; Image processing; Lesion segmentation; Machine learning; Neuroendocrine; Oncology; PPGL; Pheochromocytoma and paraganglioma.

Fig. 1

Workflow of 3D full resolution nnUNet model for automated 68GA-DOTATATE-avid PPGL lesion segmentation. The model has been trained using concatenated 3D CT, PET, and lesion mask images. The trained network has been validated by running inference using the PET and CT images from various clinical cohorts

Fig. 2

Comparison of each TP, FN, and FP lesions SUVmax (top row) for (a) PPGL Test, (b) NET (c) ONB cohorts, TTV [cm³] (middle row) for (d) PPGL Test, (e) NET, (f) ONB cohorts and TLU [cm³] (bottom row) for (d) PPGL Test, (e) NET, (f) ONB cohorts. Statistical significance was tested using the clustered Wilcox test and only the statistically significant differences (* p < 0.05) are shown with the paired line between lesion types. Without any annotations, they are not statistically significant pairwise (p > 0.05)

Fig. 3

(a) ⁶⁸GA-DOTATATE PET maximum intensity projection (MIP) image of a patient from the PPGL cohort, (b) lesion annotations by experts (c) lesion annotations by artificial intelligence (AI) model with the marking of one false positive (FP) (black dotted circle) and two false negative (FN) lesions) (yellow dotted circle) (DSC = 0.96). (d)-(e) Axial PET/CT images, showcasing two FN lesions (SUVmax of 6.62, 17.32) (marked in a yellow dotted circle) by AI

Fig. 4

(a) ⁶⁸GA-DOTATATE PET maximum intensity projection (MIP) image of a patient from the PPGL cohort, (b) lesion annotations by experts (c) AI predicted lesion annotation (DSC = 0.86)) with one false negative (FP) (yellow dotted circle) and multiple very small false positive lesions (black dotted circle). (d) Axial PET/CT images, showcasing the FN lesion by AI. (e)-(f) Axial PET/CT images, showcasing multiple FP lesions by AI marked in a yellow dotted circle. AI identified bone lesions with SUVmax in the range 5.14 to 5.42, which have been missed in the expert annotation due to using a semi-automated threshold (SUV ≥ 6) to create the annotations

Fig. 5

Histogram representation of the frequency of different types of lesions in different anatomical structures for all the cohorts