Biomedical Engineering Reference
In-Depth Information
careful. selection. of. the. parameters. within. the. algorithms. is. important. to. ensure. that. they. converge.
eiciently..In.the.case.of.the.genetic.algorithm,.for.example,.if.too.large.a.population.is.used,.the.initial.
search.space.is.too.large.and.convergence.is.slow,.similarly.if.the.mutation.value.is.incorrect,.the.search.
is.too.random.or.in.the.case.of.a.small.mutation.value.too.slow.and.only.a.small.area.of.the.optimization.
landscape.is.explored..Selecting.the.correct.acceptable.solution.is.also.important..If.too.strict.a.limit.is.
placed.on.the.convergence,.for.example,.the.inherent.noise.in.the.system.(laser.luctuations.and.detector.
noise).will.mean.that.the.system.never.reaches.a.true.plateau..his.is.a.particular.challenge.for.nonlinear.
methods.where.small.changes.in.excitation.intensity.clearly.have.a.nonlinear.efect.in.the.signal..hus,.
before.running.any.optimization.methods,.users.should.develop.an.excellent.understanding.of.the.noise.
inherent.in.their.system.
For.the.results.presented.later,.we.used.a.Simplex.optimization.routine.(Nelder.and.Mead.1965).that.had.
previously.been.used.for.astronomical.AO..In.brief,.for.a.mirror.with.
N
.actuators,.we.require.
N
+.1.initial.
mirror.shapes.and.these.are.generated.through.a.lookup.table.ater.a.series.of.experiments.to.determine.
the.best.basis.set.to.start..A.graphical.interface.is.also.used.that.enables.the.full.image.to.be.displayed.and.
a.region.of.interest.selected.for.the.optimization.algorithm.to.analyze..Options.are.provided.on.the.metric.
to.be.used.and.the.metric.value.is.displayed.in.a.real-time.graph..he.mirror.shape.is.also.displayed.with.
false.color.to.express.the.wavefront.distortion.as.a.set.of.Zernike.terms..he.system.is.capable.of.recording.
all.of.the.images.and.wavefronts.produced.for.later.analysis..One.advantage.of.the.Simplex.routine.in.this.
context.is.that.the.original.mirror.shapes.can.be.carefully.selected.to.ensure.that.initial.expected.aberra-
tions.can.be.corrected.quickly..hus,.the.standard.set.of.mirror.shapes.used.initially.speciically.contains.
shapes.to.correct.for.spherical.aberrations.and.astigmatism,.which.are.known.to.be.large.in.most.samples..
Speciically,.when.using.the.0.75.NA.air.objective.described.earlier,.we.know.that.there.will.be.spherical.
aberration.caused.when.imaging.samples.mounted.in.water.
11.3.2 results
he.results.presented.below.are.taken.from.thick.mouse.backskin.tissue.stained.with.hematoxylin.and.
eosin.at.a.depth.of.around.100.µm..Although.not.normally.used.as.a.luorescent.label,.hematoxylin.and.
eosin.provides.a.stale.sample.with.minimal.photobleaching.thus.ensuring.that.the.metrics.and.not.the.
experimental. conditions. are. being. tested.. Figure. 11.5. shows. typical. before. and. ater. images—in. this.
case,.using.the.Sobel.metric..he.gray.line.shows.the.position.used.for.intensity.cross.sections.
Ten.optimizations.were.run.for.each.metric.for.the.image.shown.in.Figure.11.5..he.average.improve-
ment. of. the. metric. value. between. the. image. before. and. ater. the. optimization. was. calculated. and. a.
standard.deviation.was.given.to.estimate.how.repeatable.the.optimization.is.
he.results.are.shown.in.
Figure.11.6
.
demonstrating.the.relative.improvement.on.an.optimized.image.
with.all.the.diferent.metric.functions..To.directly.compare.the.metrics,.once.the.image.was.optimized.
using.each.of.the.metrics,.the.improvement.made.was.tested.using.every.other.metric..he.horizontal.
axis.in.
Figure.11.6
.
shows.each.of.the.metrics.used.to.assess.the.inal.improvement..Clearly,.the.ive.sets.
of.ive.vertical.bars.all.have.very.diferent.amplitudes..However,.the.bars.within.each.block.of.ive.show.
Sobel filter (edge detection)
~
80 microns
FIGuRE 11.5
Sample.of.mouse.back.tissue.imaged.before.and.ater.optimization.of.the.adaptive.optics.using.a.
Sobel.metric.
