A general method of pattern recognition was applied to the problem of recognizing extracellularly-recorded neuronal action potentials in the presence of noise and other pulses. A PDP1 1/23 performed the calculations. There were four stages: 1.) A bandpass filter attenuated noises; 2.) the data input program digitized the signal every 55 μsec. If the signal exceeded a threshold, 12 samples of the signal and the time were written onto the disk; 3.) the pulse discriminating program recognized an action potential by fitting the 12 points with this function: v(t) = (a + bt + ct² ) exp(-t/Ƭ). For each pulse the computer determined values of the parameters giving the best fit through use of the least squares technique. For acceptance, the total pulse height and the position of the zeroes of v(t) must fall within limits; 4.) occasionally a pulse may be missed or an extra one recorded. The computer displayed the complete pulse train and the operator moved a cursor to insert or delete pulses.
Remmel, Ronald S.
"Computer Pattern Recognition of Action Potentials,"
Journal of the Arkansas Academy of Science: Vol. 37
, Article 18.
Available at: http://scholarworks.uark.edu/jaas/vol37/iss1/18