An end-to-end deep learning pipeline to derive blood input with partial volume corrections for automated parametric brain PET mapping

TL;DR

The study tackles the invasive arterial sampling requirement for dynamic brain FDG-PET by introducing an end-to-end deep learning pipeline that derives an image-derived input function from the internal carotid arteries with partial volume corrections, enabling non-invasive Ki mapping via the Patlak model. It combines a 3D U-Net–based ICA-net for automatic ICA segmentation and a hybrid time-series MCIF-net (LSTM and Bi-directional GRU) to correct for spillover and PV effects, producing a Model Corrected Input Function for downstream Ki estimation. Across 5-fold cross-validation on 50 datasets, ICA-net achieved a Dice of 82.18% and IoU of 68.54%, while MCIF-net with Bi-directional GRU + LSTM attained an MAE of 0.0526 and an MSE of 0.0052, demonstrating high accuracy in non-invasively deriving kinetic inputs. In a case study of epilepsy, the pipeline accurately localized hypometabolic regions in the left hippocampus (Z ≈ -2.03) with RMSE ≈ 0.068 against a reference model, and the resulting treatment led to a seizure-free period, underscoring the approach’s clinical relevance and potential to broaden dFDG-PET adoption.

Abstract

Dynamic 2-[18F] fluoro-2-deoxy-D-glucose positron emission tomography (dFDG-PET) for human brain imaging has considerable clinical potential, yet its utilization remains limited. A key challenge in the quantitative analysis of dFDG-PET is characterizing a patient-specific blood input function, traditionally reliant on invasive arterial blood sampling. This research introduces a novel approach employing non-invasive deep learning model-based computations from the internal carotid arteries (ICA) with partial volume (PV) corrections, thereby eliminating the need for invasive arterial sampling. We present an end-to-end pipeline incorporating a 3D U-Net based ICA-net for ICA segmentation, alongside a Recurrent Neural Network (RNN) based MCIF-net for the derivation of a model-corrected blood input function (MCIF) with PV corrections. The developed 3D U-Net and RNN was trained and validated using a 5-fold cross-validation approach on 50 human brain FDG PET datasets. The ICA-net achieved an average Dice score of 82.18% and an Intersection over Union of 68.54% across all tested scans. Furthermore, the MCIF-net exhibited a minimal root mean squared error of 0.0052. The application of this pipeline to ground truth data for dFDG-PET brain scans resulted in the precise localization of seizure onset regions, which contributed to a successful clinical outcome, with the patient achieving a seizure-free state after treatment. These results underscore the efficacy of the ICA-net and MCIF-net deep learning pipeline in learning the ICA structure's distribution and automating MCIF computation with PV corrections. This advancement marks a significant leap in non-invasive neuroimaging.

Figures (4)

• Figure 1: Abstract overview of the complete pipeline with two disjoint supervised models for segmentation and blood input correction
• Figure 2: Deep Learning Models
• Figure 3: Prediction results of ICA-net and MCIF-net on two test subjects
• Figure 4: Comprehensive End-to-End Predictive Analysis of Ground Truth Data 'E30'