ABSTRACTArtificial neural networks are being widely implemented for a range of different biomedical imaging applications.Convolutional neural networks are by far the most popular type of deep earning architecture,but often require very large datasets for robust training and evaluation We introduce deep learning diffusion fingerprinting (DLDF), which we have used to classifydiffusion-weighted magnetic resonance imaging voxels in a mouse model of glioblastoma (GL261 cell line), both prior to and in response to Temozolomide (TMZ) chemotherapy.We show that, even with limited training, DLDF can automatically segment brain tumours from normal brain, can automatically distinguish between young and older (after 9 days of growth) tumours and that DLDF can detect whether or not a tumour has been treated with chemotherapy.Our results also suggest that DLDF can detect localised changes in the underlying tumour microstructure, which are not evident using conventional measurements of the apparent diffusion coefficient (ADC).Tissue category maps generated by DLDF showed regions containing a mixture of normal brain and tumour cells, and in some cases evidence of tumour invasion across the corpus callosum, which were broadly consistent with histology.In conclusion, DLDF shows the potential for applying deep learning on a pixel-wise level,which reduces the need for vast training datasets and could easily be applied to other multi-dimensional imaging acquisitionsAbbreviationsANNartificial neural networkCTx-ray computed tomographyPETpositron emission tomographyCNNconvolutional neural networkHARDIhigh angular resolution diffusion weighted imagingNODDIneurite orientation dispersion and density imagingVERDICTvascular, extracellular and restricted diffusion for cytometry in tumoursDLDFdeep learning with diffusion fingerprintingTMZTemozolomidePFAparaformaldehydeH&Ehematoxylin and eosinGFAPglial fibrillary acidic protein