<p>Quantitative susceptibility mapping (QSM) is considered as a valuable technique to assess the evolution of iron oxide nanoparticles (IONPs)-based biomaterials by deconvolving the magnetic field through the combination of magnitude and phase map information of magnetic resonance imaging (MRI). However, issues may arise during QSM processing, particularly in tissues with irregular boundaries, where signal loss can occur. In this work, we developed an improved dilated attention gate U-Net (iDAGUNet) network, an augmentation of the U-Net segmentation framework that incorporates attention mechanisms and dilated convolutional structures. Enhancement in segmentation ability is evident in iDAGUNet model with an average intersection-over-union (IoU) of 0.931. Notably, iDAGUNet-assisted QSM could effectively resolve the signal loss issue associated with irregular boundaries as evidenced by comparison with manually labeled masks, affirming their accuracy and reliability (R²=0.996). Finally, long-term residence capabilities of IONPs-incorporated hydrogels in rat vagus nerve were monitored by iDAGUNet-assisted QSM with only 41% degradation within 16 weeks. This technology exhibits substantial promise in advancing quantitative assessment of IONPs-based biomaterials evolution for preclinical evaluations and the translation of IONPs-based biomaterials into practical applications.</p>