Studies in mathematical model observers are becoming more and more common in the field of medical imaging. Mathematical observers are sophisticated image processing methods over human observers because it does not require spending extensive time and patience consuming studies. Mathematical observers are also more consistent that human observers because they do not tend to vary performance over a long period of time.

Mathematical model observers could serve as artificial intelligence for tumor detection in the future. Numerous studies from University of Massachusetts Medical have shown there are significant positive correlations between human observer performances and mathematical model observer performances. The studies also indicate that these mathematical model observers could someday replace human observers. Currently human observers are standard for making assessments of detection performance in medical imaging.

A mathematical model observer is an algorithm or an equation that takes a multivariate input, such as an image, and returns and scalar output, such as a rating. The algorithm is generated by taking in account characterizations of the image, the background, and system noise. Detection of the specified characterizations tends to be very difficult for human observers.

The work described in the dissertation is for the CHO (Channelized Hotelling Observer) which is a linear observer. A linear observer incorporates linear functions of the image pixel values. The CHO has correlated with human observers for "signal known exactly" (SKE) tasks. SKE tasks are responsibilities in which the observer is told the tumor location. The only decision to be made for a given image is if the tumor is actually present or not.

The CHO was assembled and tested in simulations via Mathworks MATLAB. The work describes implementation and theory of the CHO. The studies were limited in terms of data sets but they were sufficient in proving the usability of the CHO. The findings of this study are a leap forward in the long process of slowly replacing mathematical model observers for tumor detection instead of human observers and making the observers more consistent and reliable.