🤖 AI Summary
To address insufficient accuracy, efficiency, and generalizability in 3D medical image segmentation, this paper proposes Bio-inspired PNN-UNet—a novel architecture mapping the neural anatomy of planarians onto deep network topology. It features a dual-pathway collaborative design (Deep-UNet and Wide-UNet) integrated with a dense-connection autoencoder serving as a global feature fusion “brain” module. The method combines 3D convolutions, multi-scale feature fusion, and optional data augmentation, enabling end-to-end training. Evaluated on 3D MRI hippocampal segmentation, it achieves statistically significant improvements in Dice score over UNet and multiple state-of-the-art variants (p < 0.01), while demonstrating superior robustness and cross-dataset generalizability. The core contribution lies in pioneering the integration of neurobiologically inspired principles—specifically modular organization and functional division-cooperation mechanisms—into the architectural design of 3D medical image segmentation networks.
📝 Abstract
Our study presents PNN-UNet as a method for constructing deep neural networks that replicate the planarian neural network (PNN) structure in the context of 3D medical image data. Planarians typically have a cerebral structure comprising two neural cords, where the cerebrum acts as a coordinator, and the neural cords serve slightly different purposes within the organism's neurological system. Accordingly, PNN-UNet comprises a Deep-UNet and a Wide-UNet as the nerve cords, with a densely connected autoencoder performing the role of the brain. This distinct architecture offers advantages over both monolithic (UNet) and modular networks (Ensemble-UNet). Our outcomes on a 3D MRI hippocampus dataset, with and without data augmentation, demonstrate that PNN-UNet outperforms the baseline UNet and several other UNet variants in image segmentation.