Kaisa Helminen-Aiforia Technologies-CEO
Kaisa Helminen,
CEO
Aiforia Technologies

Pathology is increasinglygoingdigital. Slides and samplesaredigitized, imagesareviewed online, and information is sharedthroughthecloud. Thedigitalrevolution is coupledwithautomation. It is timenow for notonlytheanalysis of thoseimages to takeplace online butalso for thisanalysis to beautomated.

Deep learning (DL), a type of artificialintelligence (AI), learnsfrom data ratherthanthroughexplicitprogramming. It is highlyflexible and scalable and an extremelypowerfultool for image analysis. TheAiforiaPlatformenableseasyaccess to deeplearning and thecreation of algorithms for theanalysis of anyhistopathological image; allowingpathologists to speeduptheirwork, makenewdiscoveries, and workwithconsistency.

Improvesaccuracy and speed

For thefirsttimeever, deeplearning AI modelsareable to mimichumans in learning to recognizecomplexvisualfeatures in image data. However, DL is faster, oftenmoreaccurate and canthereforesurpasshumancapability. It canbedeployed in a hugevariety of applications, fromobjectquantification to tissueclassificationbased on morphology, and giveaccurate, quantitativeinformationfrombiologicalsamples. Pathologistscanthereforenowautomatemanual and time-consuming image analysiswork. 

Time is givenback to focus on moreimportanttasks: decision-making, collaboratingwithpeers, and savingtimeawayfrombeinghunchedover a microscope for hours on end. Withtheconvenience of a portabletabletorotherdevicethepathologistcannowautomateanalysis and reviewthe data and examinethespecimen, anywhere, anytime. In oneresearchapplication, an Aiforiadeeplearning AI modeldetected and countedalldopaminergicneurons in rodentbrainsubstantianigra in 5 seconds, compared to thenormal 45 minuteswithtraditionalmethods, withcomparableaccuracy. 

Enablesnewdiscoveries

Deep learningcanalsofindwhatthehumaneyesometimescannotsee on itsown. Featuresthatareeithertoosmall, heterogenous in theirexpression, orfound in a hugequantityspreadover a largeareaareeasilyrecognizableby AI. Theprocess of discoverycanbeenhanced, especiallywhenconsideringthedevelopment of noveldrugmolecules, as deeplearningexcels at detectingsubtledifferencesbetweenstudygroups.

Thesmallest of changes, notpossible to visualizemanually, canbedetectedbythe AI models. Furthermore, youcantrainthemwithexternalgroundtruthsnotreadilyidentifiablefromtheimagesprovided, such as response to treatment. Predictivemodelscanbetrainedwithoutcome data. Thesethenlearn to extract and visualizethefeaturesassociatedwiththeoutcome. 

In oneresearchproject AI modelsweredevelopedwithAiforiaCreate to enable a pathologist to quantifythenumber of livercellspositive for a specificbiomarkerindicating bile ductinjury. WithoutAiforiathistaskwouldhavebeennearlyimpossible, requiringmultiple, highly-trainedpathologistsalldetecting and quantifyingcells in theexactsame, uniform manner for severalhours a day for manyweeks. Theincreasedquantitative and novel data producedbydeeplearning AI allowspathologists to betterassess and determinediagnostic and prognosticindicators and criteria, leadinghealthcaremanystepscloser to providingpersonalizedmedicine. 

Completes tasks consistently

Inter- and intra-observersubjectivity is a criticalissue in image analysisacrossdiagnosticsettings, manyresearchfields, and manysampletypes. Whether a specific feature is scored, canvaryfromone person to another, and evenwithinthesamepathologistfromoneday to thenext. Deep learning AI howeverstaysconsistent. Thealgorithmsclassifyresults and solveproblemswithconcordance, alwaysaccording to thegroundtruththeyweregiven. In onesuchresearchapplication, Aiforiawasdeployedbypathologists to countlargequantities of Th+ neurons. Over 150,000 objectsweredetected in thesamples for thisonestudy. Notonlywastheprocessspeedierthanwithtraditional, manualmethodsbutabsoluteconsistencywasachieved. 

Conclusion

Aiforia is a software platformallowingusers to createtheirowndeeplearning AI models to automatetheanalysis of a variety of tasks in a variety of histopathologicalimages, withoutanycodingor hardware needed. Deep learning AI shouldbeviewed as an assistant to thepathologist, not a replacement. A tireless, consistent, speedyassistant, here to help youdiscovermore. Find out moreabouttheAiforiaPlatformhere.