• QCG-DTI enables accurate diffusion tensor estimation from minimal DW images with flexible q-space sampling schemes, overcoming limitations of fixed sampling schemes.
• Introduces a q-space-coordinate-embedded feature consistency strategy and a QCF module to embed q-space coordinates into multiscale features, eliminating dependence on fixed diffusion sampling.
• The multiscale feature residual dense (MRD) module enhances feature extraction and image reconstruction through dual-branch convolutions with different kernel sizes.
• Achieves approximately 15% reduction in mean absolute error on fractional anisotropy and 25% on mean diffusivity compared to state-of-the-art deep learning methods.