Title: Clog Detection

Speaker: Gunnar Karlsson, Pharmacia Diagnostics

Time and Place:

Wednesday, November 14th, at 15:00
Room 1116, floor 1 at Magistern, Dag Hammarskjölds väg 31, Uppsala

Abstract:

A pipetting system used for aspiration of patient sample in the Pharmacia UniCAP-system is studied. The patient sample might contain small clots that can clog the pipette or affect the test result. The samples with clogged pipettes should be aborted and those with aspirated clots should be marked to enable further evaluation of the clot's effect on the test result.

 

The pressure inside the tubing is measured using a pressure sensor and the surface drop with a level detector. A damped spring system is simulated to identify and isolate properties of the pipetting system. Different approaches are made on the classification problem of signals from the pipetting system, the three studied in depth are:

• A Neural Network with neurons in layers is trained to perform the signal classification.
• With help from the spring model a simpler solution is made. A spring, which is extended, has an internal pressure, the same for the pipetting system when the pipette is clogged. This means the pressure inside the tubing is lower after aspiration than it was before. The volume of the aspirated sample is roughly proportional to this pressure difference.
• Most samples have approximately the same appearance. When the samples with clots are aspirated the pressure dips as the clot hits the pipette tip and then returns to normal when the clot enters the bigger cavity in the pipette. The root square sum of the difference between the normal sample and the pressure for these signals is used as a classifier.

The proposed algorithm is based on 700 measurements from patient samples with ~50 clogs and ~100 clotted samples. Around 500 other measurements were made with other fluids.