Evolving Digital Circuits for Early Stage Breast Cancer Detection Using Cartesian Genetic Programming Zahra Khalid, Gul Muhammad Khan, Arbab Masood Ahmad Abstract—Cartesian Genetic Programming (CGP) is explored to design an optimal circuit capable of early stage breast cancer detection. CGP is used to evolve simple multiplexer circuits for detection of malignancy in the Fine Needle Aspiration (FNA) samples of breast. The data set used is extracted from Wisconsins Breast Cancer Database (WBCD). A range of experiments were performed, each with different set of network parameters. The best evolved network detected malignancy with an accuracy of 99.14%, which is higher than that produced with most of the contemporary non-linear techniques that are computational expensive than the proposed system. The evolved network comprises of simple multiplexers and can be implemented easily in hardware without any further complications or inaccuracy, being the digital circuit. Keywords—Breast cancer detection, cartesian genetic programming, evolvable hardware, fine needle aspiration(FNA). I. I NTRODUCTION B REAST Cancer has always remained the center of attention for active research. Cancer is a malignant tumor, an uncontrolled growth of cells that invade the surrounding tissues at early stages and spread to other areas of the body hence moving to complex levels. This work explore a computational technique to design an optimum circuit for early stage Breast cancer detection. In the last century, massive development has been made in medical science and new treatment, remedial procedures have been introduced. For these to be effective, it is necessary to develop systems for reliable diagnostic decisions. Breast Cancer is the disease which accounts for a considerable percentage of mortality rate, of all cancer deaths, including both genders, but mostly women. Timely detection increases the chances of survival. There are several confounding factors which lead to late diagnosis of the disease. Conventionally accepted dogma for these delays is that most of the time people hesitate from undergoing painful diagnostic procedures. Failure to diagnose the disease is the second most common reason for diagnostic delays. For a pathologist, it is a routine and critically very important task to identify the presence or absence of cancer cells in patient’s samples. The fatigue and low expertise of a pathologist both can lead to wrong diagnosis. There is no doubt that diagnosis of breast cancer is a challenging and a difficult job. At the same time, it is extremely laborious to identify few malignant cells among millions of normal cells, through a microscope. The work done in this research is to assist pathologists in making more Zahra Khalid is with the University of Engineering and Technology, Pakistan (e-mail: khalid.xahra@gmail.com). accurate decisions. Fine Needle Aspiration (FNA) data was taken from Wisconsin Breast Cancer Database (WBCD) [1]. Cartesian Genetic Programming is used to classify the data. The CGP network is first trained with a large number of FNA samples and then tested with equal number of test samples, to assess system’s performance. A. Evolvable Hardware Evolvable Hardware is an area, where emphasis is placed on the utilization of evolutionary techniques to make specialized electronic circuits without needing traditional applied engineering. It is an amalgamation of reconfigurable hardware, artificial intelligence, problem tolerance and autonomous systems. It utilizes application of Evolutionary and Biologically Inspired Algorithms for the specific motive of creating novel designs of optimized physical circuits and systems. Simulators and reconfigurable hardware ensures the accuracy with respect to the final hardware designs, which might include device simulators [23], or actual devices [24]. Two main methods used to evolve the circuit, in Evolvable Hardware are: Extrinsic and Intrinsic evolution. 1) Extrinsic Evolution: Extrinsic Evolution is characterized by the assessment of electronic circuits through simulation instead real-time construction during training and testing phases, and later on the final design is evaluated on real hardware. Success of this technique manifestly depends upon two important components of the system, the simulation software program used and the kind of digital additives within the final hardware implementation. Thus when implemented and evaluated in hardware, it cannot be ensured that the extrinsically evolved circuit would work as predicted. The fundamental problem with extrinsic evolvable hardware is that if elementary components like simple AND, OR and NOT gates are used, simulation of the system becomes easy and time taken to develop the circuit is manageable depending on the computer. But as soon the problem grows to be more complicated and the additives end up more complex, the time required for the simulation increases considerably. After sufficient evolution cycles a satisfactory performing circuit is produced. The final design if digital usually shows exactly same behaviour as its simulation, for analogue systems it can be tricky. 2) Intrinsic Evolution: Intrinsic Evolution is characterized by in-circuit evolution of the hardware instead of using simulation models. The circuit’s fitness is evaluated at run-time World Academy of Science, Engineering and Technology International Journal of Electrical and Information Engineering Vol:12, No:9, 2018 651 International Scholarly and Scientific Research & Innovation 12(9) 2018 ISNI:0000000091950263 Open Science Index, Electrical and Information Engineering Vol:12, No:9, 2018 publications.waset.org/10009546/pdf