Three-dimensional U-Net with transfer learning improves automated whole brain delineation from MRI brain scans of rats, mice, and monkeys

Abstract

Background: Automated whole-brain delineation (WBD) techniques often struggle to generalize across pre-clinical studies due to variations in animal models, magnetic resonance imaging (MRI) scanners, and tissue contrasts. We developed a 3D U-Net neural network for WBD pre-trained on organophosphate intoxication (OPI) rat brain MRI scans. We used transfer learning (TL) to adapt this OPI-pretrained network to other animal models: rat model of Alzheimer's disease (AD), mouse model of tetramethylenedisulfotetramine (TETS) intoxication, and titi monkey model of social bonding.

Methods: We assessed an OPI-pretrained 3D U-Net across animal models under three conditions: (1) direct application to each dataset; (2) utilizing TL; and (3) training disease-specific U-Net models. For each condition, training dataset size (TDS) was optimized, and output WBDs were compared to manual segmentations for accuracy.

Results: The OPI-pretrained 3D U-Net (TDS = 100) achieved the best accuracy [median[min-max]] for the test OPI dataset with a Dice coefficient (DC) = [0.987 [0.977-0.992]] and Hausdorff distance (HD) = [0.86 [0.55-1.27]]mm. TL improved generalization across all models [AD (TDS = 40): DC = 0.987 [0.977-0.992] and HD = 0.72 [0.54-1.00]mm; TETS (TDS = 10): DC = 0.992 [0.984-0.993] and HD = 0.40 [0.31-0.50]mm; Monkey (TDS = 8): DC = 0.977 [0.968-0.979] and HD = 3.03 [2.19-3.91]mm], showing performance comparable to disease-specific networks.

Conclusions: The OPI-pretrained 3D U-Net with TL achieved accuracy comparable to disease-specific networks with reduced training data (TDS ≤ 40 scans) across all models. Future work will focus on developing a multi-region delineation pipeline for pre-clinical MRI brain data, utilizing the proposed WBD as an initial step.

Keywords: Brain segmentation; Convolutional neural network; Magnetic resonance imaging; Preclinical models; Transfer learning.