key: cord-0153745-4qszyzak authors: Saborit-Torres, J. M.; Saenz-Gamboa, J. J.; Montell, J. A.; Salinas, J. M.; G'omez, J. A.; Stefan, I.; Caparr'os, M.; Garc'ia-Garc'ia, F.; Domenech, J.; Manj'on, J. V.; Rojas, G.; Pertusa, A.; Bustos, A.; Gonz'alez, G.; Galant, J.; Iglesia-Vay'a, M. de la title: Medical imaging data structure extended to multiple modalities and anatomical regions date: 2020-09-30 journal: nan DOI: nan sha: ee7a20ab545097e2d802454495f64b9d1560b8d2 doc_id: 153745 cord_uid: 4qszyzak Brain Imaging Data Structure (BIDS) allows the user to organise brain imaging data into a clear and easy standard directory structure. BIDS is widely supported by the scientific community and is considered a powerful standard for management. The original BIDS is limited to images or data related to the brain. Medical Imaging Data Structure (MIDS) was therefore conceived with the objective of extending this methodology to other anatomical regions and other types of imaging systems in these areas. Methods which yield reliable and reproducible results must be used when acquiring scientific knowledge. High test-retest reliability of the applied methods is the foundation of research, irrespective of the scientific discipline. It is in the prime interest of every scientict to obtain reproducible results. While such reproducibility was considered of utmost importance in the positron emitting tomography (PET) field [1] , the quantitative assessment of reproducibility has largely been neglected in the fMRI community, or as Bennett and Miller described it: "Reliability is not a typical topic of conversation" between functional magnetic resonance imaging (fMRI) investigators [2] . This situation changed significantly in 2016 following the establishment of the Committee on Best Practices in Data Analysis and Sharing [3] by the leading neuroimaging society -the Organisation for Human Brain Mapping (OHBM). The basis of the Valencia Medical Imaging Bank [4] is the clinical environment data curation proposal, by which imaging data can be collected correctly and efficiently. Finding a way to organise this information is crucial. A proper organization and curation of the images is essential to train deep learning methods [5] that can perform object detection and segmentation using reliable medical data. Metadata can also be included in multimodal classifiers to complement imaging data in order to improve the accuracy of the detection. Images and medical information can be stored in different ways, although there is no standard that indicates how this information should be organised and shared. The Health Ministry's Centre of Excellence and Innovation for Image Technology recommends using a simple system so that any researcher can understand the data distribution [6] . The proposed structure is called MIDS (Medical Imaging Data Structure) and aims to be a new standard that contains all types of medical information and images in simple hierarchical folders. It was conceived as an extension of the standard Brain Imaging Data Structure [7, 8] , which stores brain images. MIDS takes this system further and is not confined to brain images only. The idea is to create the same structure for images of different body parts by magnetic resonance, computed tomography, ecography, etc, following a single process, regardless of the type and shape of the image. Many studies focus on obtaining a medical imaging dataset for their own purposes, so that the management and control of the associated images and metadata can be roughly an effort. During projects, more data are generated and it may be necessary to relocate it inside a dataset. Each study has its own manner of organising the data, which makes it more difficult to understand, while a curated and well structured dataset can improve the search user experience and the quality of automatic classifiers. There are a couple of studies which propose a standard to store this type of data, including BIDS, which aims a standard form of storing magnetic resonance imaging data and metadata in a clear and simple hierarchical folder structure. It is supported by several programs and libraries dedicated to the study of medical images (e.g. c-pacs, freesurfer, XNAT, BIDS Validator, among others) and is widely used by research groups. Figure 1 gives an example of the BIDS structure; the left directory is a folder with DICOM (Digital Imaging and Communication On Medicine) images [9] and the right directory is its corresponding BIDS structure. As BIDS only supports brain MRIs, if a project needs, for example, a lumbar MRI, BIDS would not support the images. However, by expanding its structure, other imaging techniques can be integrated in it, which is how MIDS was created. BIDS is thus a potential standard to store MRIs and there is in practice little difference between BIDS and MIDS. Furthermore, in epidemiological studies based on Population Image, MIDS can incorporate any type of [10] medical image (e.g. Computed Radiography, Computed Tomography, Ultrasound, Mammography, etc). MIDS can thus be seen as an extension of BIDS with a similar structural format [11] . MIDS adds a new level to the BIDS directory hierarchy which describes the types of medical images used for a particular session. As can be seen in Figure 2 , the structure is compatible with BIDS. The added level is to define the type of medical imaging and can be classified by the energy used for their acquisition, together with the functional or tomographic adjectives for their generation. The classification is shown in Table 1 . This template includes a new level to describe other types of medical image than MRI. The researcher decides whether or not to use particular filename keys, depending on the type of medical image. For example, the contrast enhancement (ce-