Elż­bi­eta Trypka

Title

Screen­ing for ear­ly onset of demen­tia using voice as biomarker 

Abstract

Intro­duc­tion: We present a screen­ing method for ear­ly demen­tia using fea­tures based on sound objects as voice biomarkers. 

Meth­ods: The final dataset used for machine learn­ing mod­els con­sist­ed of 266 obser­va­tions, with a dis­tri­b­u­tion of 186 healthy indi­vid­u­als, 46 diag­nosed with Alzheimer’s, and 34 with MCI. This method is based on six-sec­ond record­ings of the sus­tained vow­el /​a/​spo­ken by the sub­ject. The main orig­i­nal con­tri­bu­tion of this work is the use of care­ful­ly craft­ed fea­tures based on sound objects. This approach allows one to first rep­re­sent the sound spec­trum in a more accu­rate way than the stan­dard spec­trum, and then build inter­pretable fea­tures con­tain­ing rel­e­vant infor­ma­tion about sub­jects’ con­trol over their voice. 

Results: ROC AUC obtained in this work for dis­tin­guish­ing healthy sub­jects from those with MCI was 0.85, while accu­ra­cy was 0.76. For dis­tin­guish­ing between healthy sub­jects and those with either MCI or Alzheimer’s the results were 0.84, 0.77, respectively. 

Con­clu­sion: The use of fea­tures based on sound objects enables screen­ing for ear­ly demen­tia even on very short record­ings of lan­guage-inde­pen­dent voice samples. 

Biog­ra­phy

Elż­bi­eta Tryp­ka, MD, PhD, is an Assis­tant Pro­fes­sor at the Depart­ment of Psy­chi­a­try at the Wro­claw Med­ical Uni­ver­si­ty. She is a spe­cial­ist in psy­chi­a­try, a co-founder, and a board mem­ber of the Pol­ish Psy­chogeri­atric Society. 

Her main research inter­ests focus on neu­rode­gen­er­a­tive dis­eases of the cen­tral ner­vous sys­tem, ear­ly diag­nos­tic pos­si­bil­i­ties, and exist­ing bio­chem­i­cal dis­or­ders in the course of Alzheimer’s dis­ease. She is the author and co-author of sci­en­tif­ic papers pub­lished in Pol­ish and inter­na­tion­al jour­nals. Notably she serves as the par­tic­i­pant and leader of the many projects relat­ed to research seek­ing ear­ly bio­mark­ers of demen­tia dis­or­ders. She leads pio­neer­ing research on Sound Objects is Vivid Mind’s unique, patent­ed approach­to audio decom­po­si­tion, sep­a­ra­tion and analy­sis using sound vec­tor­ing meth­ods and algo­rithms. This advanced sig­nal pro­cess­ing method­ol­o­gy is based on a sinu­soidal mod­el that offers a rad­i­cal­ly new approach to com­pu­ta­tion­al analy­sis of the audi­to­ry scene.