An automated medical trend detection program, TrenDx, was developed in earlier work by Haimowitz and Le. It was evaluated on its ability to discern growth abnormalities by matching templates of expected growth patterns. The results of these evaluations were somewhat disappointing because the program was inefficient and could not reach the level of sensitivity and specificity of human physicians in referral decisions. This thesis involved engineering improvements in the original program and evaluations of the new updated program. The engineering changes allow the program to run on faster machines and eliminate the long run times. The revised scoring algorithms effectively prevent the program from reaching erroneous conclusions from too little data. Re-evaluation of previous data and analysis of newly collected data show genuine improvements in the performance of TrenDx, which now performs at a level comparable to physicians and may soon be used in a clinical setting.
Submitted May 22, 2000 to the Department of Electrical Engineering and Computer Science in partial fulfillment of the requirements for the degree of Master of Engineering in Electrical Engineering and Computer Science.
Thesis Supervisor: Peter Szolovits, Professor of Electrical Engineering and Computer Science