O'Reilly's notes on 10 am meeting.

Attendees: Heidi Sosik, Denis Klimov, Tom O'Reilly

Questions:
Plans for small/low-power instrument?
Can FCB/IFCB be deployed without power/data cable-to-shore?
HS: Working on "low-relief" IFCBT for deployment on Wave Glider, AUV, other platform
Current IFCB takes 30W power

Particle size range overlap of FCB and IFCB?
HS: Difficult; IFCB laser is optimized for microplankton - picoplankton requires higher-power laser. FCB is still original prototype (large) - no funding or plans to refine. CytoBuoy claims pico-micro capability.

Could you replace sheath fluid with virtual core technology, ala SeaFlow (Need to add position-sensitive detector(s))?
HS: Would be concerned about flow-cell cleanliness - core fluid keeps particles moving through flow cell. Very occassionally get dirt in cell - would this be a greater problem with SeaFlow/SeaLabel?

Processing:
HS: Several stages; detect particles (reject particle-free images regions). Edge detection, etc. Classification
Now using "random forest" approach instead of SVM (2007):
- rf requires less tuning than svm
- svm requires complete characterized dataset for error characterization; rf doesn't
- both methods yield comparable accuracy

How long did the image-processing/classification procedures take per image? On what hardware?
HS: particle detection/empty-image rejection executes in situ on instrument's "Atom" processor. Shore-side workflow in realtime, posted on MVCO website (http://ifcb-data.whoi.edu/)
HS: Image processing is more computationally intensive than classification

Klimov's notes from 10:00, 12:00, 16:00 meetings

Principles of operation:
-       FlowCytobot / Imaging FlowCytobot sample path arranged in such a way as sample goes "down hill" due to plankton sinking.
-       Newer Imaging FlowCytobot (licensed to Mclane) is developed entirely at WHOI. McLane did some adaptations to simplify production. 8 inch diameter tube, 40 inch long. First unit built in August 2013.
-       5ml sip, then 20 minutes analyzing. This time can be reduced by sipping less, say 1ml then 4 min analyzing, but the period can not be much shorter due to time overhead (pump start/ stop etc) and required statistics.
-       this design has emphasis on accurate statistics so flow and volume need to be precise, so syringe pump is used with defined volume.
-       Emphasis on "binning" to get particle statistics per "shot" and not as continuous measurement; so "binning" is ok here.
-       Also, accurate flow rate is required, due to separation of "trigger volume" and image acquisition volume within a sample cuvette, and fixed time delay between trigger and acquisition of the image. Image is acquired with 1 us light pulse from Xenon lamp.
-       Sheath fluid is preventing the contact between sample water and walls of the cuvette, minimizes chance of contamination and particles attachment to cuvette walls, very important feature.
-       Sheath fluid is recirculated by filtering sample + sheath by volumetric Fisher filter. Excess water is dumped into environment before filter. Filter lasts 6-8 month deployment easy.
 
Use of Imaging FlowCytobot on a mobile platform:
-       Configuration of the sample path dictates "vertical" position of the instrument so it operation in horizontal orientation could be problematic.
-       For glider use, perhaps the Imaging FlowCytobot can be split into set of housings and then arrange them horizontally.
-       Suggestion: keep volume of those compartments relatively small (under several hundred mL) to avoid vehicle loss in case of flooding one compartment.
 
 
Difference between FlowCytobot and Imaging FlowCytobot:
-       Particle ranges: FlowCytobot 1-15 um, Imaging FlowCytobot 5-200um
-       Cuvettes have different geometries: FlowCytobot has smaller cylindrical flow cavity for tighter light collimation of laser, and Imaging FlowCytobot has "50 um sheet" of fluid optimized for imaging and larger laser beam for triggering
-    Light sources are optimized differently for smaller circular path, or wider flat path