The ever increasing interest in evolutionary algorithms (EAs) has been reflected in the increasing numbers of publications and shows no sign of declining. The evaluation of EAs reported in these papers is often achieved by analysing their performance at solving one or more mathematical problems such as linear regression. However, use of EAs in real-world applications has also been increasing. The area of clinical medicine is no exception and it is here that developments are particularly interesting. This Special Issue of Genetic Programming and Evolving Machines is intended to give an update of the advances in this area and has been motivated by the successful GECCO Workshop series on Medical Applications of Genetic and Evolutionary Computation (MedGEC), which has its fourth annual event in Atlanta, USA in 2008. The purpose of the Workshop is two-fold—to provide a venue to report the latest developments in a theoretically demanding and life-critical area and to facilitate a forum in which practitioners from scientific, engineering and clinical backgrounds can discuss, criticise and support work in this important field.

In this Special Issue seven papers have been included which are representative of the following areas: optimisation, modelling, discovery and design and classification.

In An Evolutionary Approach to Cancer Chemotherapy Scheduling by Gabriela Ochoa, Minaya Villasana and Edmund Burke, the design of a patient’s chemotherapy schedule is considered an optimal control problem. A mathematical model of tumour growth is used to determine effective drug schedules that eradicate the tumour whilst ensuring that the chemotherapy side-effects are maintained at an acceptable level. Using a single-objective function with several terms, a de-randomised evolutionary strategy with covariance matrix adaptation is employed which was found to perform better than a conventional genetic algorithm and simulated annealing algorithm with various operators.

Another optimisation problem is considered in the second paper, Interactive Evolution for Cochlear Implants Fitting by Pierrick Legrand et al., but this time also incorporates learning and an interactive element. A cochlear implant is used to provide hearing sensation to patients with severe to profound hearing loss for which a conventional hearing aid is no longer effective. The implant comprises a number of electrodes, which are surgically attached to different parts of the cochlear providing stimulation at different frequencies in response to sound conveyed by microphone and associated digital signal processing equipment. The configuration of the equipment has many parameters, is highly specific to a particular patient, and cannot be effectively modelled. The process is also complicated by other factors such as acute patient fatigue. The paper describes an approach to optimising the parameters of the implant using an interactive evolutionary algorithm with a micro-population. Results suggest that a useful optimisation of the parameters has been achieved by using the evolutionary algorithm, but also that the traditional strategy adopted in clinical practice might not be ideal and that patient fatigue is a particularly important consideration when evaluating implant quality.

Optimisation can also play an important role in modelling and this is exemplified in the paper Stochastic Optimization of a Biologically Plausible Spino-neuromuscular System Model—A Comparison with Human Subjects by Stanley Gotshall et al. The spino-neuromuscular system, the means by which the brain communicates with muscles, is largely understood in terms of its functionality and the major signal pathways, but less so in terms of the control processes employed. This paper describes the use of genetic algorithms and a particle swarm optimizer towards a model that may ultimately significantly benefit the diagnosis of disease and treatment of injury to the spino-neuromuscular system. Interestingly, the results also show that the model can spontaneously adopt biologically plausible behaviours at both neural and gross anatomical levels without direct selective pressure.

A second modelling-based paper, Using Evolvable Genetic Cellular Automata to Model Breast Cancer by Armand Bankhead and Robert Heckendorn is based on the premise that cancer is an evolutionary process in which mutated cells may be selected for abnormal growth, sometimes resulting in a tumour. A common form of breast cancer, ductal carcinoma in situ, is modelled using cellular automata and examines hereditary predisposition on early incidence and aggressiveness of the cancer. Results not only demonstrate these pathological features, but also that the progenitor hierarchy structure plays a major role its increased incidence and aggressiveness of the cancer.

The following two papers both use evolutionary approaches in discovery and design. In Genomic Mining for Complex Disease Traits with “Random Chemistry,” Margaret J. Eppstein et al. propose data mining of the human genome to determine the genetic etiology of common diseases which have complex traits resulting from multiple gene–gene and gene–environment interactions. Specifically, an evolutionary hill-climbing approach, inspired by Kauffman’s random chemistry is proposed. The success of the algorithm is heavily dependent on the definition of a suitable fitness function based on an approximate and noisy fitness function. Results show that the technique is capable of detecting small DNA sequence variations (single nucleotide polymorphisms) indicative of disease.

The second discovery and design paper, Genetic Programming for Computational Pharmacokinetics in Drug Discovery and Development by Francesco Archetti et al. considers the effectiveness of genetic programming when compared with other machine learning techniques in predictive pharmacokinetics, or drug development. Specifically, a genetic program outperformed other techniques in predicting two important characteristics of a drug: oral bioavailability and median oral lethal dose. Although linear regression and support vector machines regression performed better in prediction of the characteristic of plasma protein binding, the work demonstrates the importance of genetic programming in this rapidly advancing field.

The final paper considers a growing importance of the application of evolutionary algorithms to classification. In Diagnosis of Parkinson’s Disease using Evolutionary Algorithms by Stephen Smith et al., patients’ responses to a simple figure-copying task are digitized in real-time using a commercially available graphics tablet. The resulting stream of x–y coordinates are then pre-processed before being presented to a cartesian genetic program, a form of genetic programme in which the nodes are held in a constrained rectangular cartesian array. The results indicate the power of applying raw data to an evolutionary algorithm as opposed to merely optimising features extracted from the data in a conventional classification system.

These papers are only a selection of the wide range of subject areas submitted. Although it was pleasing to have a large response to the call for papers, which confirmed the motivation for this Special Issue, only seven could be accommodated here. The Guest Editors would like to thank all authors for submitting their work and the referees for undertaking the reviews.