Ex vivo 9.4T MRI of rectal cancer specimens after neoadjuvant therapy: structural and microstructural imaging dataset

Ex vivo 9.4T MRI of rectal cancer specimens after neoadjuvant therapy: structural and microstructural imaging dataset This dataset includes high-resolution (9.4 T) MRI of ex vivo total mesorectal excision (TME) surgical specimens from rectal cancer patients scanned after neoadjuvant therapy (NAT). The data were acquired to investigate whether high-resolution ex vivo diffusion MRI (dMRI) can characterize the properties of rectal wall tissue components - distinguishing residual tumour from treatment-induced fibrosis - more effectively than conventional T2-weighted imaging. Associated publication The acquisition and processing methods are described in: Fouto, A. R., Calá, H., Moreira, S., Shemesh, N., Fernandez, L., Couto, N., Herrando, I., Nougaret, S., Popita, R., Brito, J., Ouro, S., Chambel, M., Papanikolaou, N., Parvaiz, A., Heald, R. J., Castillo-Martin, M., Santiago, I., & Ianus, A. (2026). High-resolution advanced diffusion MRI of rectal cancer surgical specimens: correlating microstructural characteristics with histology. medRxiv. https://doi.org/10.64898/2026.04.02.26350055 Data description This dataset contains MRI data from five ex vivo rectal cancer specimens. For each specimen, the corresponding data are stored in a separate folder (sub-specimen001, sub-specimen002, ... sub-specimen005). Unzip the compressed restage-ex_vivo_v1.zip. This will extract the folder restage-ex_vivo_v1 containing version 1 of the ex vivo MRI data. Inside, you will find sub-folders sub-specimen001, sub-specimen002, ... sub-specimen005. Each specimen folder is organized into two sub-folders, anat and dwi: anat/ - structural MRI · *_T2_multi-echo.nii.gz: multi-slice multi-echo (MSME) structural MRI volume (8 echoes) · *_T2_multi-echo.json: file with the corresponding acquisition parameters dwi/ - diffusion MRI · *_dwi.nii.gz: diffusion-weighted MRI volume (4D) · *_dwi.bval: b-values used in the diffusion acquisition (one per volume; b = 1500 and 3000 s/mm²) · *_dwi.bvec: gradient directions corresponding to each diffusion volume (15 directions) · *_dwi.json: file with the corresponding acquisition parameters The "derivatives" folder contains manually-drawn tissue masks (regions of interest) defined on selected MRI slices, used for the radiology-histology correlation analysis described in the associated publication. Masks are provided per slice, since each mask corresponds to an MRI slice that was matched to a histological section. Each mask is a binary NIfTI file (.nii.gz) co-registered with the corresponding specimen's diffusion MRI volume. derivatives/ | |-- sub-specimen001/ | |-- sub-specimen001_mask_fibrosis_sliceNN.nii.gz Binary mask of fibrosis on slice NN. | |-- sub-specimen001_mask_<tissue>_sliceNN.nii.gz [Other tissue masks on slice NN] | |-- sub-specimen003/ | |-- ... | (and so on for the remaining specimens) Mask naming convention: sub-specimen<ID>_mask_<tissue>_slice<NN>.nii.gz Where <tissue> identifies the tissue type [e.g. fibrosis, tumour, mucosa, submucosa, muscle] and <NN> is the slice number in the corresponding diffusion MRI volume. Acquisition All data were acquired on a 9.4 T Bruker BioSpec preclinical MRI system (86 mm Tx/Rx coil) at 22 °C. Specimens were fixed in formalin (36 h) followed by PBS (4 h) and mounted in Fomblin for scanning. Fat suppression was applied to both sequences. Diffusion MRI (2D): TR/TE = 11,000/24 ms; 130 slices; matrix 140 × 130; isotropic resolution 0.5 mm³; 2 b0 images; b = 1500 and 3000 s/mm²; 15 diffusion directions. Structural MRI (MSME): TR = 25,000 ms; 8 echoes (TE = 10–80 ms); same geometric parameters as the diffusion acquisition. One specimen (specimen001) was acquired with a different MSME protocol: 6 echoes (TE = 15–90 ms; echo spacing 15 ms). Acknowledgements We would like to thank the Champalimaud Foundation Biobank (CFB) for their work with the pseudonymization of the surgical specimens and FFPE blocks; we would like to thank the grossing technicians of the Service of Anatomic Pathology of the Champalimaud Clinical Centre (CCC), specially João Leonardo Fernandes for generating the sections needed to do the path-rad correlation; we would like to thank the Histology Platform of Champalimaud Research for providing the tissue sections to perform the dual stainings; we would like to thank Raquel Graça-Lopes, MSc, and Marco Pereira of the Service of Anatomic Pathology of the CCC for the digitalization of the histological slides. This work was supported by the Fundação para a Ciência e a Tecnologia (2023.02201.RESTART; THCS/0003/2023) within the framework of the co-funded Transforming Health and Care Systems (THCS) partnership (GA N° 101095654) under the EU Horizon Europe Research and Innovation Programme. AI is also supported by the ERC Horizon Europe Research and Innovation Programme Starting Grant No. 101164674.