AI Tracking Solutions LLC — Department of Health and Human Services STTR Phase I: 400

Description

Project Summary AbstractParticle trackingPTis a biophysical tool for elucidating molecular interactionstransport phenomena of diverse speciesand rheological properties of complex materialsPT experiments involve first obtaining high resolution videos that capture time resolved increments of particlesfollowed by extraction of traces of entities of interest from videos in the form of spatial locations over timea process we refer to as path conversionFinallyquantitative analysis of the traces will yield diffusivitiesviscoelasticityetcLung diseasessuch as cystic fibrosis and COPDare characterized by a highly viscoelastic mucus layer that is incapable of being cleared by mucociliary clearanceNot surprisinglythe viscoelasticity of mucus often directly reflects disease progressionA variety of mucolytics are being investigatedbut due to the variable composition and properties of mucus between patientseffective mucolytics treatment will likely be different between individualstoo little inappropriate mucolytics will not be effective in restoring mucus clearancewhereas too much may result in bronchorrheaAlthough microbeads based rheology has been performed on a variety of mucus specimens in basic researchthe capacity for high throughput characterization of rheological properties of biological specimens in a clinical setting is currently not availableThis limitation can be attributed to inefficiencies of path conversioncurrent PT software requires extensive human supervision intervention to achieve accurate path conversionnot only resulting in poor reproducibility and throughput but also restricting its use to only expert labsOur vision is to make PT as objective and easy to use as a simple plate reader that can be readily utilized by cliniciansdiagnosticsdisease progressiontherapy effectivenesspharmapreclinical clinical drug screeningand research professionalsTowards this goalwe have created a neural network trackerNNTthat automatically determines the location of all particles in each frame with zero user inputi eno parameter for users to changeand retains the identity of all particles from frame to frameThe innovation is that NNT can robustlyreproduciblyand accurately track a wide range ofDD videos with virtually no need for human interventionachieving unparalleled time savingsWe have already successfully deployed NNT over the Google cloudwhich offers exceptional scalabilityNeverthelessfor time sensitive applicationssuch as an automated PT rheometerthe transfer of large video data files is likely prohibitiveThereforein this Phase I STTRwe seek to enable real time NNT based PT analysis on the local machine while video microscopy data is being acquired by the microscopeand allow data from PT analysis to drive the operation of the microscopeIn Aimwe will integrate our NNT with a single objective fluorescence microscope system called MonoptesAimwill evaluate the performance of our NNTMonoptes systemIf successfulour technology would form the basis of a fully automated PT system capable of measuring rheological properties of fluids materials or distribution of particle sizes in awell plate format Project Narrative Particle tracking is a powerful biophysical tool in life and physical sciencesbut unfortunatelyits application has been strongly limited by inefficiencies in accurately extracting particle traces from raw moviesUnlike conventional particle tracking methodswe have combined artificial intelligence and machine learning to create a software that can consistently provide superior and truly automated tracking performance compared to current alternativesIn this proposalwe will integrate this latest advance with sophisticated instrumentation to develop a microscope system capable of fully automated particle tracking microscopy in awell plate formatIf successfulthe instrument will likely be utilized by cliniciansdiagnosticsdisease progressiontherapy effectivenesspharmapreclinical clinical drug screening of patientsand research professionals