{"id":4535,"date":"2025-05-14T17:33:02","date_gmt":"2025-05-14T15:33:02","guid":{"rendered":"https:\/\/datalchemy.net\/nos-prestations\/ai-expertise-support\/"},"modified":"2025-08-19T10:18:39","modified_gmt":"2025-08-19T08:18:39","slug":"ai-expertise-support","status":"publish","type":"projet","link":"https:\/\/datalchemy.net\/en\/our-services\/ai-expertise-support\/","title":{"rendered":"AI expertise support"},"content":{"rendered":"\n

Context<\/strong><\/h2>\n\n

Datalchemy supports Chipiron, an innovative start-up whose mission is “to put an end to cancer thanks to scalable MRI technologies”.<\/p>\n\n

Today, access to MRI – the most accurate, safe and non-invasive imaging modality – remains limited by its high cost, cumbersome infrastructure and hospital compatibility constraints. To enable widespread screening and image-guided therapies, accessibility needs to be multiplied by 100: Chipiron is therefore developing the first full-body MRI scanner with ultra-low field strength (< 10 mT), which is lightweight, mobile (one person only), operates on a standard socket and can be used without a shielded room, without sacrificing image quality. <\/p>\n\n

Support work<\/strong><\/h2>\n\n

We set up a targeted scientific watch on AI advances applied to MRI methods, then selected and evaluated each publication according to four key axes:<\/p>\n\n

    \n
  1. General confidence<\/strong> (1 low – 5 high): reputation of authors and laboratories, number of citations, rigor of data and clarity of results.<\/li>\n\n\n\n
  2. Reproduction and maintenance effort<\/strong> (1 high – 5 low): availability of codebases, checkpoints and datasets, quality of documentation, responses to issues, clear separation of train\/test and ablation studies.<\/li>\n\n\n\n
  3. Generalizability to acquisition parameters<\/strong> (1 low – 5 high): robustness of results to protocol variations, theoretical arguments for independence from acquisition conditions.<\/li>\n\n\n\n
  4. Expected performance<\/strong> (1 low – 5 high): computational cost for training and inference, image quality obtained.<\/li>\n<\/ol>\n\n

    Each article selected is the subject of a summary sheet with figures and comments, together with an overall “industrializability” score and recommendations for prioritizing the work to be prototyped.<\/p>\n\n

    Results<\/strong><\/h2>\n\n

    Thanks to this rigorous, ongoing analysis, Chipiron was able to demonstrate to its investors the scientific soundness and technical maturity of its project. This directly contributed to the closing of a $17 million Series A round at the end of April 2025, backed by Blast Club, IXCORE Group, the European Innovation Council (EIC) and Bpifrance Chipiron<\/a>. This financing will enable : <\/p>\n\n