Multi-channel orthogonal convolutional neural networks

1. A method of improving discrimination between superficial lipid and calcium versus fibrous tissue and lipid, calcium tissues, and connective tissue, the method comprising:

(1) creating a database of a-scans characteristic of each fibrous, calcium, lipid, and connective tissue based on histology and user input;

(2) parsing individual a-scans one at a time from a b-scans;

(3) delimiting a tissue region;

(4) identifying an index of an initiation of a signal decay region;

(5) identifying an index of a termination of the signal decay region;

(6) calculating a goodness-of-fit (GOF) to a Gaussian function;

(7) extracting a denominator coefficient in the Gaussian function;

(8) calculating an area under a signal decay region;

(9) calculating an area under a total delimited tissue region; and

(10) inputting statistics from steps (4) and (5) into a linear discrimination analysis (LDA) trained on the database to classify an a-scan as fibrous, calcium or lipid.

2. The method of claim 1 further comprising biasing thresholds on a neural network based on a-scan classification obtained in step (10) of claim 1 .

3. The method of claim 1 wherein delimiting a tissue region comprises sampling from a start of a lumen to a point where an intensity is five percent of a maximum intensity.

4. The method of claim 1 wherein identifying an index of an initiation of a signal decay region comprises:

using a panning window algorithm where slope is calculated between intensity values at end points of a window; and

determining a signal decay region when five consecutive windows show a negative slope.

5. The method of claim 1 wherein identifying an index of a termination of the signal decay region comprises identifying five consecutive windows with positive slope in the signal decay region.