The M/EDGE pro­ject aims to de­velop and pro­to­type an elec­tron­ics plat­form for highly in­te­grated med­ical edge in­tel­li­gence for multi-sen­sor cap­sule en­doscopy and neu­roim­plants.

Mod­ern med­ical de­vices, with their tight in­te­gra­tion of pro­gram­ma­ble mi­cro­elec­tron­ics, sen­sors, and ac­tu­a­tors, have opened fun­da­men­tally new di­ag­nos­tic and ther­a­peu­tic pos­si­bil­i­ties. The log­i­cal evo­lu­tion to cloud-net­worked cy­ber-med­ical sys­tems of­fers rev­o­lu­tion­ary op­por­tu­ni­ties for in­tel­li­gent, care­ful, and pin­pointed med­i­cine by using AI tech­niques. How­ever, for med­ical im­plants or wire­less sen­sor tech­nol­ogy, broad­band and un­in­ter­rupted net­work con­nec­tiv­ity is tech­ni­cally and prac­ti­cally un­fea­si­ble. The so­lu­tion to these chal­lenges is the in­te­gra­tion of in­tel­li­gence and au­ton­omy di­rectly in the med­ical de­vice, i.e. med­ical edge com­put­ing. For the de­ploy­ment of ma­chine learn­ing in such cy­ber-med­ical edge de­vices, there is a high de­mand for high-per­for­mance em­bed­ded hard­ware ar­chi­tec­tures that can ex­e­cute ma­chine learn­ing for in­tel­li­gent sen­sor data pro­cess­ing in an en­ergy-ef­fi­cient man­ner with­out vi­o­lat­ing ap­pli­ca­tion-spe­cific re­quire­ments in terms of la­tency and power con­sump­tion. An im­por­tant pre­req­ui­site for this goal is the pro­vi­sion of in­tel­li­gent edge com­po­nents that can be em­bed­ded in their en­vi­ron­ment in a well-tai­lored man­ner and in­ter­act with it au­tonomously. This re­quires ap­pli­ca­tion-spe­cific AI hard­ware ac­cel­er­a­tors that can be flex­i­bly adapted to the spe­cific re­quire­ments of med­ical ap­pli­ca­tions for the analy­sis and clas­si­fi­ca­tion of sen­sor data streams. In this con­text, the ma­chine learn­ing meth­ods are to be de­vel­oped and op­ti­mized to­gether with the hard­ware AI ac­cel­er­a­tors in an au­to­mated HW/SW code­sign in order to meet the re­quired per­for­mance char­ac­ter­is­tics. This en­ables ef­fi­cient in­tel­li­gent sen­sor sig­nal pro­cess­ing with an av­er­age elec­tri­cal power con­sump­tion well below 1 mW.

Fund­ing

The pro­ject M/EDGE is funded by the Fed­eral Min­istry of Ed­u­ca­tion and Re­search (BMBF).

Par­tic­i­pat­ing Team Mem­bers

Bring­mann, Oliver

Gerum, Christoph

Reiber, Moritz

Werner, Julia

Bause, Oliver