AI-based medi­cal devices: MDR ver­sus AI Regulation

The inte­gra­ti­on of arti­fi­ci­al intel­li­gence (AI) has beco­me a must-have for a wide ran­ge of medi­cal devices. AI sup­ports dia­gno­stics and offers decis­i­on sup­port for the­ra­pies and medi­ca­ti­on to medi­cal pro­fes­sio­nals. Medi­cal apps often only achie­ve full func­tion­a­li­ty through the use of AI.

For manu­fac­tu­r­ers of AI-based medi­cal devices, the ques­ti­on ari­ses as to which regu­la­to­ry requi­re­ments their medi­cal devices must com­ply with if they want to use AI, and how the regu­la­to­ry frame­works of Regu­la­ti­on (EU) 2017/745 on medi­cal devices (MDR) and Regu­la­ti­on (EU) 2017/746 on in vitro dia­gno­stic medi­cal devices (IVDR) in con­junc­tion with the future Euro­pean Regu­la­ti­on estab­li­shing har­mo­nis­ed rules for arti­fi­ci­al intel­li­gence (AI Regu­la­ti­on), curr­ent­ly available as a final draft (draft AI Regu­la­ti­on), are to be applied.

No clear definition

Neither MDR nor IVDR include a defi­ni­ti­on of the term “arti­fi­ci­al intel­li­gence” or “AI.” The­re is also no clear defi­ni­ti­on of the term “soft­ware”. In prin­ci­ple, howe­ver, soft­ware con­sti­tu­tes a medi­cal device or IVD under the MDR and IVDR if it is inten­ded by the manu­fac­tu­rer to ful­fill a spe­ci­fic medi­cal purpose.

The draft AI Regu­la­ti­on, on the other hand, defi­nes the term “arti­fi­ci­al intel­li­gence sys­tem” or “AI sys­tem”, for short, as “soft­ware that is deve­lo­ped with one or more of the tech­ni­ques and approa­ches lis­ted in Annex I [of the draft AI Regu­la­ti­on] and can, for a given set of human-defined objec­ti­ves, gene­ra­te out­puts such as con­tent, pre­dic­tions, recom­men­da­ti­ons, or decis­i­ons influen­cing the envi­ron­ments they inter­act with.” Annex I of the draft AI Regu­la­ti­on in turn descri­bes tech­ni­ques of and approa­ches to AI, in par­ti­cu­lar approa­ches to machi­ne lear­ning inclu­ding deep lear­ning, logic and knowledge-based approa­ches as well as sta­tis­ti­cal approa­ches and esti­ma­ti­on, search and opti­mi­sa­ti­on methods.

Inter­sec­tion: AI-based medi­cal devices as high-risk AI systems

Over­laps bet­ween the Regu­la­ti­ons ari­se in the case of AI-based medi­cal devices, i.e. for soft­ware that, accor­ding to the manufacturer’s deter­mi­na­ti­on, ser­ves the ful­fill­ment of a medi­cal pur­po­se and was deve­lo­ped with one of the approa­ches and tech­ni­ques desi­gna­ted in Annex I of the draft AI Regu­la­ti­on in order to achie­ve cer­tain objec­ti­ves. With regard to AI-based medi­cal devices, both the MDR/IVDR and the AI Regu­la­ti­on are fun­da­men­tal­ly rele­vant, espe­ci­al­ly sin­ce a lar­ge num­ber of AI-based medi­cal devices in the terms of the MDR/IVDR will at the same time have to be qua­li­fied as “high-risk AI sys­tems” in the terms of the AI Regulation:

In accordance with Artic­le 6(1) of the draft AI Regu­la­ti­on, “high-risk AI sys­tems” are AI sys­tems that them­sel­ves fall as a pro­duct under a har­mo­ni­sa­ti­on pro­vi­si­on lis­ted in Annex II of the AI Regu­la­ti­on or that repre­sent a “secu­ri­ty com­po­nent” of a pro­duct that its­elf falls under the scope of appli­ca­ti­on of a regu­la­ti­on lis­ted in Annex II. Ano­ther pre­re­qui­si­te is that a “third-party con­for­mi­ty assess­ment” is spe­ci­fied in the respec­ti­ve case in accordance with the pro­vi­si­ons men­tio­ned the­re. The MDR and IVDR con­sti­tu­te har­mo­ni­sa­ti­on regu­la­ti­ons encom­pas­sed by Annex II. Thus, if the AI-based medi­cal device is to be assi­gned to risk class IIa or hig­her accor­ding to Rule 11, and if the invol­vement of a Noti­fied Body in the con­for­mi­ty assess­ment pro­ce­du­re is man­da­to­ry, the AI-based medi­cal device or the soft­ware as a medi­cal device also repres­ents a high-risk AI sys­tem in the terms of the draft AI Regulation.

Spe­ci­fic regu­la­to­ry requi­re­ments for high-risk AI systems

The draft AI Regu­la­ti­on descri­bes a ple­tho­ra of regu­la­to­ry requi­re­ments that will need to be met by manu­fac­tu­r­ers of AI-based medi­cal devices as well as pro­vi­ders and users, and importers, dis­tri­bu­tors and aut­ho­ri­sed repre­sen­ta­ti­ves. Chap­ter 2 of the draft AI Regu­la­ti­on con­ta­ins, among other things, essen­ti­al requi­re­ments regarding

  • risk manage­ment,
  • data gover­nan­ce,
  • tech­ni­cal docu­men­ta­ti­on and recor­ding obligations,
  • trans­pa­ren­cy and pro­vi­si­on of infor­ma­ti­on to users,
  • accu­ra­cy, robust­ness, and cyber secu­ri­ty for high-risk AI systems.

What is striking here is that the MDR and IVDR also alre­a­dy pro­vi­de for detail­ed requi­re­ments on most of the­se aspects. Main­tai­ning tech­ni­cal docu­men­ta­ti­on or imple­men­ting qua­li­ty or risk manage­ment sys­tems is not new. Howe­ver, in order to meet the requi­re­ments of the future AI Regu­la­ti­on as a manu­fac­tu­rer of an AI-based medi­cal device, it is neces­sa­ry to “think along” with the spe­cial aspects of qua­li­ty assu­rance of the AI com­po­nent or the eva­lua­ti­on of the spe­ci­fic risk poten­ti­al in the con­text of a con­for­mi­ty assess­ment and to include them from deve­lo­p­ment through manu­fac­tu­ring and product/user infor­ma­ti­on to pro­duct monitoring.

Recom­men­da­ti­on for action for manu­fac­tu­r­ers of AI-based medi­cal devices

  • Ensu­re that the AI tech­no­lo­gy used com­pli­es with MDR or IVDR requi­re­ments for software.
  • Learn about the requi­re­ments of the AI Regu­la­ti­on for high-risk AI sys­tems that go bey­ond the requi­re­ments of the MDR and IVDR.
  • Con­duct com­pre­hen­si­ve risk assess­ments to ensu­re that the use of AI in your medi­cal devices is safe and does not cau­se adver­se effects.
  • Add expl­ana­ti­ons to the tech­ni­cal docu­men­ta­ti­on on the use of AI in your medi­cal devices, inclu­ding the exact natu­re of the tech­no­lo­gy used and its per­for­mance characteristics.
  • Work clo­se­ly with accre­di­ted Noti­fied Bodies or Noti­fied Bodies that have the exper­ti­se to gui­de the con­for­mi­ty assess­ment of AI-based medi­cal devices.
  • Stay abre­ast of the latest deve­lo­p­ments regar­ding AI and medi­cal devices – espe­ci­al­ly with regard to the evo­lu­ti­on of the lia­bi­li­ty régime – and adapt your pro­ducts and pro­ces­ses accordingly.

Stay up-to-date

We use your email address exclusively for sending our newsletter. You have the right to revoke your consent at any time with effect for the future. For further information, please refer to our privacy policy.