Purpose: The use of Artificial Intelligence (AI) for personalized medicine has recently guided dramatic improvements in the diagnostic pathway of several diseases. The European Project GENOMED4ALL, aims at using European level data of patients affected by Multiple Myeloma, Myelodysplastic Syndromes and Sickle Cell Disease (SCD) to find correlation between genomics – and other omics data – with phenotypic manifestations and seize the opportunity of improving diagnostics through AI. Silent Cerebral Infarcts (SCIs) are a significant determinant of morbidity since childhood in SCD. One of the aims of the SCD clinical case in GENOMED4ALL is the use of radiomics – quantitative method for the evaluation and interpretation of medical images- and AI firstly to develop an automatic and uniform identification and characterization of SCI through the analysis of cerebral MRI, secondly, to correlate imaging data with other types of omics data in order to predict risk of recurrence. Materials and methods: The MRI protocol included 3D-T1, FLAIR and DWI sequences. Neuroradiological reports were cross checked for consistency. A stepwise segmentation (identification of lesion volume), pre-processing and extraction of MRIs was performed utilizing different open-source software: Lesion Segmentation Tools and UNet for segmentation, Freesurfer for brain extraction. To optimize SCI identification, we have selected the best software for segmentation; freesurfer was substituted with a faster approach. Results: Six European SCD centers participated in this first phase of the Radiomics project with 501 MRIs: 225 were classified as abnormal by the local neuroradiologists due to presence of SCI, 275 were normal; 70% were pediatric MRIs. Different instruments were used in the 6 centers: Philips 1.5 T (n.2), Siemens 1.5 T and 3 T (n.2), GE 3T (n.2). A stepwise procedure allowed the optimization of SCI identification, with distinction between SCI and background non clinically significant noise (i.e periventricular areas) or other white matter hyperintensities (i.e transient glial maturation). As shown in Figure 1, to segment SCI in FLAIR MRI, we have used a pre-trained UNet [Li H, 2018]. Firstly, we have extracted the brain registering the MNI152 on the T1 image. Then we have applied the brain mask and a threshold taking only the largest component. The UNet performs the segmentation of hyperintensities. We have removed the regions surrounded by less than the 90% of White Matter or near the brain ventricles to remove non-SCI region. In the end, we have enlarged the remaining ones. To date, segmentation on 303 exams from different centers showed very few false positive and false negatives highlighting the need to take into account different technical characteristics due to the various equipment used in the different centers. Conclusion: SCD is a systemic disorder with extreme phenotypic variability. Radiomics and AI offer the opportunity to seize the potential of big datasets analysis to optimize diagnostic in SCD. SCI can be detected automatically from different datasets. Four more European Centers will add their MRI in the second phase of the project, to increase variability and allow correlation of in detailed Radiomics data (lesion volume, lesion site, etc) with clinical-hematological characteristics and other omics data.Figure 1. Stepwise Segmentation (identification and selection of lesion volume) workflow for automatic selection of all SCI. From left to right: Axial slices of the input scan, Slice after skull stripping to highlight only the brain parenchyma, Segmentatio The authors do not declare any conflict of interest