Next: Analyzing typing speed, speech and sleep patterns to identify cognitive decline, dementia, Parkinson’s, and more

by

AI May Help Iden­ti­fy Patients With Ear­ly-Stage Demen­tia (The Wall Street Journal):

Researchers are study­ing whether arti­fi­cial-intel­li­gence tools that ana­lyze things like typ­ing speed, sleep pat­terns and speech can be used to help clin­i­cians bet­ter iden­ti­fy patients with ear­ly-stage dementia.

Huge quan­ti­ties of data reflect­ing our abil­i­ty to think and process infor­ma­tion are now wide­ly avail­able, thanks to watch­es and phones that track move­ment and heart rate, as well as tablets, com­put­ers and vir­tu­al assis­tants such as Ama­zon Echo that can record the way we type, search the inter­net and pay bills…

The goal of using arti­fi­cial intel­li­gence in health care isn’t to replace humans but rather to assist doc­tors, says P. Murali Doraiswamy, pro­fes­sor and direc­tor of the Neu­rocog­ni­tive Dis­or­ders Pro­gram at Duke Uni­ver­si­ty School of Med­i­cine. “This isn’t a bat­tle between AI and doc­tors, it’s about how to opti­mize doc­tors’ abil­i­ty to deliv­er bet­ter care,” he says.

Dr. Doraiswamy has col­lab­o­rat­ed on sev­er­al projects involv­ing machine learn­ing and neu­rode­gen­er­a­tive dis­eases, includ­ing a study of inter­net-search behav­ior with Microsoft Corp. In that study, researchers found that machine-learn­ing algo­rithms trained to ana­lyze sub­jects’ cur­sor move­ments in terms of speed, changes in direc­tion and tremors, as well as whether the sub­jects repeat­ed search queries or repeat­ed­ly clicked on search results, could help detect Parkinson’s dis­ease. Pre­lim­i­nary analy­ses showed the strat­e­gy holds promise for detect­ing Alzheimer’s, as well.

The Study:

Detect­ing neu­rode­gen­er­a­tive dis­or­ders from web search sig­nals (npj Dig­i­tal Medicine):

  • Abstract: Neu­rode­gen­er­a­tive dis­or­ders, such as Parkinson’s dis­ease (PD) and Alzheimer’s dis­ease (AD), are impor­tant pub­lic health prob­lems war­rant­i­ng ear­ly detec­tion. We trained machine-learned clas­si­fiers on the lon­gi­tu­di­nal search logs of 31,321,773 search engine users to auto­mat­i­cal­ly detect neu­rode­gen­er­a­tive dis­or­ders. Sev­er­al dig­i­tal phe­no­types with high dis­crim­i­na­to­ry weights for detect­ing these dis­or­ders are iden­ti­fied. Clas­si­fi­er sen­si­tiv­i­ties for PD detec­tion are 94.2/83.1/42.0/34.6% at false pos­i­tive rates (FPRs) of 20/10/1/0.1%, respec­tive­ly. Pre­lim­i­nary analy­sis shows sim­i­lar per­for­mance for AD detec­tion. Sub­ject to fur­ther refine­ment of accu­ra­cy and repro­ducibil­i­ty, these find­ings show the promise of web search dig­i­tal phe­no­types as adjunc­tive screen­ing tools for neu­rode­gen­er­a­tive disorders.

News in Context: