review Article Oman Medical Journal [2015], Vol. 30, No.

6: 406–416

The Applications of Genetic Algorithms in Medicine

Ali Ghaheri1, Saeed Shoar2, Mohammad Naderan3 and Sayed Shahabuddin Hoseini4*
1
Department of Management and Economy, Science and Research Branch, Azad University, Tehran, Iran
2
Department of Surgery, Shariati Hospital, Tehran University of Medical Sciences, Tehran, Iran
3
School of Medicine Tehran University of Medical Sciences, Tehran, Iran
4
Hannover Medical School, Germany

A RT I C L E I N F O A B S T R AC T
Article history: A great wealth of information is hidden amid medical research data that in some cases
Received: 8 August 2015 cannot be easily analyzed, if at all, using classical statistical methods. Inspired by nature,
Accepted: 21 October 2015 metaheuristic algorithms have been developed to offer optimal or near-optimal solutions
Online: to complex data analysis and decision-making tasks in a reasonable time. Due to their
DOI 10.5001/omj.2015.82 powerful features, metaheuristic algorithms have frequently been used in other fields of
sciences. In medicine, however, the use of these algorithms are not known by physicians
Keywords:
Medicine; Algorithms; who may well benefit by applying them to solve complex medical problems. Therefore, in
Genetic Algorithm; this paper, we introduce the genetic algorithm and its applications in medicine. The use of
Optimization; Heuristic the genetic algorithm has promising implications in various medical specialties including
methods. radiology, radiotherapy, oncology, pediatrics, cardiology, endocrinology, surgery, obstetrics
and gynecology, pulmonology, infectious diseases, orthopedics, rehabilitation medicine,
neurology, pharmacotherapy, and health care management. This review introduces the
applications of the genetic algorithm in disease screening, diagnosis, treatment planning,
pharmacovigilance, prognosis, and health care management, and enables physicians to
envision possible applications of this metaheuristic method in their medical career.

T
here is no doubt that computers have dynamics (based on the process of river formation),7
revolutionized our everyday life. They artificial immune systems (based on immune system
are vastly used and have benefited nearly function),8 and genetic algorithm (inspired by
all fields of science from aerospace genetic mechanisms).9 Metaheuristic approaches
and astronomy to biology, chemistry, physics, have been frequently used in other fields of science
mathematics, geography, archeology, engineering, where complex problems need to be solved, or
and social sciences. optimal decisions should be made. In medicine,
In medicine, electronic chips and computers although valuable work has been done, the power of
are the backbones of a lot of imaging, diagnostic, these potent algorithms for offering solutions to the
monitoring, and therapeutic devices. These devices, countless complex problems physicians encounter
which are composed of several different hardware every day has not been fully exploited.
components, are managed and controlled by In this paper, we introduce the genetic algorithm
software, which in turn are based on algorithms. (GA) as one of these metaheuristics and review some
An algorithm is a set of well-described rules and of its applications in medicine.
instructions that define a sequence of operations.
Metaheuristic methods are algorithms that can more The genetic algorithm
quickly solve complex problems, or they can find an A GA is a metaheuristic method, inspired by the laws
approximate solution when classical methods are not of genetics, trying to find useful solutions to complex
able to find an exact one.1 problems. In this method, first some random
Several metaheuristic algorithms for finding solutions (individuals) are generated each containing
an optimal or near-optimal solution exist. These several properties (chromosomes). Based on the
include the ant colony (inspired by ants behavior),2 laws of genetics, cross-over and mutations occur in
artificial bee colony (based on bees behavior),3 Grey chromosomes to produce a second generation of
Wolf Optimizer (inspired by grey wolves behavior),4 individuals with more diverse properties.
artificial neural networks (derived from the neural Crossover and mutation are the two most central
systems),5 simulated annealing,6 river formation methods for diversifying individuals. In crossover,

*Corresponding author: [email protected]

 A li Gh aheri, et al. 407

a
First Generation

Crossover

Second Generation

Figure 1: Methods to induce diversity in the population of individuals (candidate solutions). (a) During
crossover, one part of a chromosome is exchanged by another fragment of another chromosome. (b) During
mutations, one or more datasets on a chromosome are converted to different ones. These alterations will
generate new individuals whose fittest (more optimal solutions) will survive.

two chromosomes are chosen. Then a crossover point based methods search only a single point. Moreover,
along each chromosome is chosen followed by the GAs select the next population using probabilistic
exchange of the values up to the crossover point transition rules and random number generators
between the two chromosomes [Figure 1]. These while derivative-based algorithms use deterministic
two newly-generated chromosomes produce new transition rules for selecting the next point in the
offspring. The process of crossover will be iterated sequence.11,12
over and over until the desired diversity of individuals In the following, we introduce some of the
(i.e. solutions) is made. The mutation also generates applications of GAs in a variety of medical disciplines.
new configurations by applying random changes
in different chromosomes.10 One of the simplest Radiology
mutation methods has been depicted in Figure 1. Imaging techniques in radiology generate a large
In a GA, the possibility of reproduction depends amount of data that needs to be analyzed and
on the fitness of individuals. The better chromosomes interpreted by radiologists in a relatively short time.
they have (i.e., those with better characteristics), the Computer-aided detection and diagnosis are rapidly
more likely they are to be selected for breeding the growing interdisciplinary technologies that aim to
next generation. There are several selection methods; assist radiologists in faster and more accurate image
however, the aim of all is to assign fitness values to analysis by detection, segmentation, and classification
individuals based on a fitness function and to select of normal and pathological patterns found on various
the fittest. Genetic alterations in chromosomes will imaging modalities. These include X-rays, magnetic
happen via crossover and mutations to produce resonance imaging (MRI), compute tomography
another generation. This iterative process will (CT) scan, and ultrasound.13
continue until the fittest individual (the optimal In machine vision, an image of scenery (such as
solution) is formed or the maximum number of organs of the human body in radiology images) is
generations is reached.9,11 acquired, processed, and interpreted. The boundaries
It is worth noting that GAs are different from the (shape) and sizes of objects within the images need
derivative-based, optimization algorithms. First of to be determined to assess the objects in more detail.
all, GAs search a population of points in the solution Therefore, the process of edge detection becomes one
space in each iteration while classical derivative- of the integral parts of automatic image processing

O M A N M E D J, V O L 3 0 , N O 6 , N O V E M B E R 2015
408 A li Gh aheri, et al.

techniques.14 Several researchers have used the to differentiate benign and malignant breast tumors
GAs for edge detection of images acquired using in ultrasound images.25
different imaging modalities including MRI, CT, de Carvalho Filho et al,28 developed a GA for
and ultrasound.14-16 automatic detection and classification of solitary
Screening mammography is the gold standard lung nodules. The designed algorithm could detect
for detection of breast cancer; however, due to lung nodules with about 86% sensitivity, 98%
its failure rate,17,18 researchers have tried to apply specificity, and 98% accuracy.
computational tools to improve the sensitivity of the Image registration or fusion is the process of
system. In fact, the majority of the applications of optimal aligning of two or more images into one
GAs in radiology were performed on breast cancer coordinate system. Precise integration of images
screening primarily using mammography. becomes crucial when valuable information is
Karnan and Thangavel19 applied the GA to detect embedded within several images acquired under
microcalcifications in mammograms suggesting of different conditions (viewpoint, sensor, or time).29
breast cancer. In their method, after enhancement GAs have successfully been used to align MRI and
and normalization of the mammograms, the border CT scan images in several studies.30,31 In another
of breast and the nipple position was detected by the study, positron emission tomography (PET) images
GA. Using the border and the nipple position of the were fused with MRI images by a GA to generate
right and left breasts as a reference, the mammogram colored breast cancer images.32
images were aligned and subtracted from each other Precise tumor staging is an important part of
to find the asymmetry image suggestive of breast designing a treatment plan. Accurate tumor size and
cancer. The Az value, which is the area under the volume determination using non-invasive imaging
receiver operating characteristic (ROC) curve, studies becomes essential for tumor staging. Zhou
has been used as a useful measure for assessing the et al,33 developed a system for extraction of tongue
diagnostic performance of a system.20 The Az value carcinoma from head and neck MRIs. A GA was
for their proposed algorithm was about 0.9.19 applied for segmentation of images followed by an
In another study, Pereira et al,21 applied a set of artificial neural network (ANN)-based symmetry-
computational tools for mammogram segmentation detection algorithm to reduce the number of false
to improve the detection of breast cancer. An positive results. This approach was able to extract
algorithm was first designed to eliminate artifacts tongue carcinoma from an MRI with high accuracy
followed by denoising and image enhancement. and minimal user-dependency.
Consecutively, combining wavelet analysis and
the GA allowed detection and segmentation of Oncology
suspicious areas with 95% sensitivity. GAs have Screening tests offer a valuable opportunity for
also been successfully used for classification and early cancer detection, which if followed by proper
detection of clustered microcalcifications in digital treatment could improve the survival rate of patients.
mammograms.22-24 To develop a non-invasive technique for cervical
In machine learning, feature selection is the cancer detection, Duraipandian et al,34 acquired
process of selecting a subset of relevant features to Raman spectra from the cervical area via colposcopy.
construct a model by removing variables with little The biomolecular information generated via
or no analytical value. Feature selection is important the Raman spectroscopy was analyzed by a GA-
since choosing irrelevant features would increase partial least square-discriminant analysis system to
the time, cost, and complexity of computation and differentiate between a normal and dysplastic cervix.
reduce the accuracy of the model.25 Besides, reducing Partial least square (PLS) is a statistical method
the number of features would avoid the problem aiming to find a linear regression model between a
of over-fitting, reduce the chance of failure upon dependent variable and some predictor variables.35
missing data, and allow for a better explanation and This system was able to differentiate dysplasia from
generalization of the model.26 a normal cervix with 72% sensitivity and 90%
GAs have been applied for feature selection in specificity.34
studies aiming to identify a region of interest in The advent of DNA microarrays has paved the
mammograms as normal or containing a mass,27 and way for massive gene expression profiling that could
 A li Gh aheri, et al. 409

revolutionize the field of molecular diagnostics for parameter estimation necessary for accurate
and prognosis. However, generation of large sets elasticity quantification to determine tissue elasticity.
of data poses statistical and analytical challenges This system is superior to gradient-based methods
necessitating the need to find key predictive genes.36 used for parameter estimation of the inefficiency
Due to the inherent capability of GAs to search and of gradient-based techniques for inhomogeneous
find the optimal solution among large and complex solution spaces containing several local minima and
possible solutions with multiple simultaneous requirement for substantial computational time
interactions, they have been applied to analyze limits their application.41
microarray data from several cancer cell lines.36 The field of biomarker discovery and clinical
Dolled-Filhart et al,37 generated microarray data by proteomics is rapidly growing in medical diagnosis,
staining breast cancer tissues with several antibodies prognosis, and disease follow-up. Advanced
specific for various markers to find a minimum set technologies such as mass spectrometry can generate
of biomarkers with maximum classification and readouts of thousands of proteins from patient
prognostication values in breast cancer patients. The samples; however, the cost and complexity of such
data analyzed using GAs showed that three markers techniques on the one hand and computational
with available antibodies could define a population and statistical methods for analysis, on the other
of patients with more than a 95% five-year survival hand, necessitates the selection of a few, relevant
rate. markers for clinical assay development. Zhou et al,42
Tan et al,38 conducted a study to investigate employed an improved version of the GA supported
the relationship between soil trace elements and by a recursive local floating enhancement technique
cervical cancer mortality in China. A combination to predict the risk of a major adverse cardiac event
of GA and PLS was used to choose five out of 25 (MACE). This technique was able to select a panel
trace elements. Then a least square support vector of seven proteins including myeloperoxidase to
machine (LSSVM) model was developed. LSSVM predict the risk of MACE with 77% accuracy, which
is a method used in machine learning to infer a outperformed over several current methods.
function from or find a pattern in training data.39 Logistic regression models have been frequently
The results showed that a combination of GA-PLS used in diagnosing diseases. Due to its outstanding
and LSSVM could predict the mortality of cervical performance, a GA has been used to select the best
cancer based on trace elements.38 variables for a logistic regression system aiming to
One of the important and informative factors model the presence of myocardial infarction in
influencing the choice of an appropriate therapeutic patients with chest pain. The GA-based method was
approach for cancer patients is determination of the superior in variable selection to other traditional
disease prognosis. In a retrospective study on more methods.26
than 200 patients, Bozcuk et al,40 compared the One of the key elements in the automatic
performance of four different data mining methods interpretation of the electrocardiogram (ECG) is
to determine the outcome of cancer patients not the detection of QRS complexes that would allow
being in terminal stages after hospitalization. In assessment of heart rate variability and other relevant
comparison to other methods, GA selected the diagnostic parameters. Tu et al,43 introduced a simple
least number of explanatory variables (lactate and effective GA to detect QRS complexes. Then,
dehydrogenase and the reason for admission) to p-waves and f-waves, which happen in normal
predict the outcome of patients. ECG and after atrial fibrillation, respectively, were
successfully extracted from patient databases. Such
Cardiology algorithms could allow comprehensive research into
GAs have been used in different fields of ECG details.
cardiovascular medicine. Atherosclerotic plaques
are hallmarks of most myocardial infarctions and Endocrinology
strokes. Determination of plaque mechanical Hypoglycemia is the most common complication
properties such as elasticity would enable physicians of insulin therapy in patients with type 1 diabetes
to locate better and map vulnerable or unstable mellitus (T1DM). Hypoglycemia can induce
plaques. Khalil et al,41 used a system involving GAs alterations in the patterns of electroencephalograms

O M A N M E D J, V O L 3 0 , N O 6 , N O V E M B E R 2015
410 A li Gh aheri, et al.

(EEGs). Nguyen et al,44 combined ANNs, GAs, and statistics was applied for data analysis. A GA was
Levenberg-Marquardt (LM) training techniques to used to select the suitable wavelets (variables) of the
detect hypoglycemia based on EEG signals. ANN electropherogram to minimize the computation time
was used to model the relationship between blood required for the Bayesian computation. This system
glucose and EEG signals. For training ANN, the was able to differentiate LGA from AGA using only
global search ability of GA and the local search two wavelets, one of albumin and the other of a
capability of LM were combined. Data from four negatively-charged unknown small molecule with
EEG parameters derived from two EEG channels 100% sensitivity and 98% specificity.48
were used by the analyzing system to detect The prediction of fetal weight before delivery
hypoglycemia with 75% sensitivity and 60% can reduce the potential problems associated with
specificity. In another paper, a GA-based multiple low-birth-weight infants. Yu et al,49 introduced fuzzy
regression with fuzzy inference system was developed logic into the support vector regression (FSVR) to
to detect non-invasive episodes of nocturnal estimate the fetal weight. GAs were used to generate
hypoglycemia in children with T1DM. Using heart an evolutionary FSVR to select the optimal features
rate and corrected QT interval, hypoglycemia was for the FSVR system. This outperformed a back-
detected with a sensitivity of 75% and specificity of propagation neural network by achieving the
over 50%.45 lowest mean absolute percent error (6.6%) and the
highest correlation coefficient (0.902) between the
Obstetrics and gynecology estimated and the actual fetal birth weight.
The differentiation between normal and prolonged
delivery allows obstetricians to determine the Pediatrics
optimal timing for interventions, if necessary, Cardiotocography is a cheap and non-invasive
during childbirth. One of the parameters that can technique to assess the fetal heart rate and uterine
help to forecast the delivery time and segregate contractions to determine fetal well-being. Ocak50
normal versus prolonged labor is the time to reach applied a GA to select the optimal features of
full cervical dilation. Hoh et al,46 applied a three- cardiotocogram recordings for a support vector
parameter logistic model using GA or the Newtone- machine (SVM) classifier. The results showed that
Raphson (NR) method to predict the time to reach the new system classified fetal health status as normal
full cervical dilation. The GA-based algorithm or abnormal with 99.3% and 100% accuracy, which
outperformed the NR method by more accurately was superior to an ANN algorithm designed for the
predicting the time to full cervical dilation. same purpose.
A Pap smear is a cytology test for detection of Autism is a neurodevelopmental disease that
precancerous and cancerous cervical changes. In this appears in early childhood and is characterized by
method, 20 features of cells are assessed to describe impaired social functioning and verbal and non-
them as normal or abnormal or, more specifically, verbal communications and repetitive behavior.
categorize them into seven classes. Marinakis et al,47 To recognize autism based on the microarray gene
generated a hybrid model that took advantage of expression data, Latkowski and Osowski51 used
the feature-selection capability of GAs to reduce GAs to select the most relevant genes associated
the complexity of features necessary for a nearest with the disease. Frequently selected genes include
neighbor algorithm for classification of Pap smear RMI1, NRIP1, TOP1, ZFHX3, CEP350, NFYA,
results. The new method outperformed several other PSENEN, ANP32A, SEMA4C, and SP1. These
previously used approaches by accurately classifying genes provided an input for an ensemble of classifiers
the Pap smear results. including SVM and random forest classifiers. The
GAs have also been applied in prenatal diagnosis. introduced system recognized autism with 96%
One of the fetal features that can complicate sensitivity and 83% specificity.51
delivery is fetal macrosomia. In an attempt to Acute lymphoblastic leukemia (ALL) is the most
differentiate the large-for-gestational-age (LGA) common type of leukemia in children and has many
from the appropriate-for-gestational-age (AGA) subtypes. Analysis of gene expression data derived
infants, amniotic fluid from the second trimester from tumor cells can help classifying cancers. Due
was evaluated by capillary electrophoresis. Bayesian to the enormous size of information generated from
 A li Gh aheri, et al. 411

microarray gene expression profiling, Lin et al,52 used diagnostic method that can differentiate lung
a GA to select the most relevant genes needed for diseases and guide the diagnostic approach toward
ALL classification. Silhouette statistics was applied more specific techniques. To automate lung sound
as a discriminant function to differentiate between diagnosis, a hybrid GANN was designed. The
six ALL subtypes. The proposed technique reached GA was applied to optimize the ANN training
a 100% classification accuracy and used fewer parameters and reduce the computation time. The
discriminating genes compared to other methods. new system could classify the lung sounds into
Aneuploidy is a condition where one or a few normal, wheeze, and crackle.56
chromosomes in the nucleus of a cell are above or Assessment of the partial pressure of carbon
below the normal chromosomal number of a species. dioxide in the arterial blood (PaCO2) is important
Conventional chromosomal studies on amniocentesis in the management of critically ill patients. To avoid
samples are performed for definite diagnosis of fetal difficulties associated with arterial blood sampling,
aneuploidy yet the rather long required time for non-invasive methods for predicting PaCO2 such
these techniques necessitates the development of as assessment of exhaled carbon dioxide at end-
faster diagnostic tests. To this end, the proteomic expiration (PetCO2) could be applied in normal
profile of the amniotic fluid specimens was identified individuals; however, their use in sicker persons might
via mass spectrometry and the generated data was be biased and less helpful. Engoren et al,57 designed
assessed by a GA. The proposed method could a GA to predict the PaCO2 using 11 variables from
detect aneuploidy with 100% sensitivity, 72%–96% capnography of non-intubated patients in the
specificity, 11%–50% positive predictive value and emergency department. The proposed system could
100% negative predictive value.53 improve the precision and bias of PaCO2 prediction.

Surgery Infectious diseases

ANNs are powerful mathematical algorithms Tuberculosis is a possible lethal infectious disease
capable of predicting the behavior of systems. not only in developing countries but also in
Due to the predictive value of ANNs, a GA-based developed nations after the emergence of human
ANN (GANN) was developed to predict the immunodeficiency virus (HIV). To predict the
outcomes after surgery for patients with non-small diagnosis (tuberculosis vs. non-tuberculosis
cell lung cancer (NSCLC). The GA was applied patients), 38 parameters composed of examination
to help optimization not to fall into local minima. parameters and laboratory data were used to design
The GANN model could predict the outcome of an ANN trained by a GA. The classification accuracy
NSCLC patients more accurately and significantly of the system was about 95%, which was higher than
better than logistic regression. Besides, the inclusion the results obtained by other algorithms.58
of tumor size in calculations significantly improved Highly active antiretroviral therapy (HAART),
prediction outcomes.54 an integral part of the treatment modalities against
As populations age, the number of geriatric HIV, is composed of a combination of several
patients needing cardiac surgeries increases. Due antiretroviral medications aiming to decrease the
to the high prevalence of comorbid conditions in replication of the virus. Since long-term HAART
elderly, proper prognostication of postoperative treatment needs patient compliance and might
morbidity and mortality would be informative, be associated with some side effects, structured
precluding overestimation of risk and denial of treatment interruption has been proposed to reduce
surgery for patients deserving it, which could not only side effects, but also the selection pressure
happen with some prediction models. Applying a on the virus that could lead to the emergence of
GA, Lee et al,55 showed that a short length of stay resistant particles. Therefore, Castiglione et al,59
after cardiac surgery was correlated with younger age, devised a GA-based system to choose the best
no preoperative use of beta blockers, shorter cross- HAART treatment schedule to control HIV and
clamp time, and absence of congestive heart failure. help the immune system to reconstitute. A virtual
model of the immune system was used to assess the
Pulmonology effects of anti-HIV drugs on virtual patients.59,60
In pulmonology, auscultation is the most common The new structured interruption schedule could

O M A N M E D J, V O L 3 0 , N O 6 , N O V E M B E R 2015
412 A li Gh aheri, et al.

achieve therapeutic results and protection against an or loosening.69 In another report, a combination
opportunistic infection comparable to a full-length of ANNs and GAs was applied to design spinal
treatment.61 pedicle screws used for fixation of spinal fractures.
The hybrid algorithm was able to design screws with
Radiotherapy a higher fatigue life and ideal pullout and bending
Intensity modulated radiotherapy (IMRT) was characteristics.70
developed to transfer an accurate dose of radiation Scoliosis is a three-dimensional deformity of
to a target such as the brain, prostate, or head and spinal axis curves. The progression of the disease,
neck. Planning IMRT involves selection of 5–10 which only happens in a small percentage of
angles for wavelet projection and determining the patients, is monitored by serial X-rays over time.
radiation dose. The application of GA could improve Since frequent exposure to X-rays might increase the
the selection of gantry angles in a reasonable time chance of cancer, it is desirable to assess the disease
frame.62 Similar GA-based irradiation planning has development using harmless methods. Jaremko
been applied for patients with other types of cancer et al,71 developed a GA-based ANN algorithm to
including pancreatic,63 rhabdomyosarcoma, and estimate the angle of spinal axis deformity from
brain tumors.64 GAs have also been successfully used indices of trunk surface deformity. The hybrid system
to optimize the design of stereotactic radiosurgery, was able to determine the angle deformity within 5%
and radiotherapy treatment plans.65 accuracy in more than two third of patients.

Rehabilitation medicine Neurology

As the need for physical rehabilitation increases, Multiple sclerosis (MS) is a debilitating inflammatory
novel treatment equipment and techniques have to disease of the neural system characterized by the
be developed and tested. Refinement of these new formation of white matter scars otherwise known
methods needs changing various parameters and as plaques. Computer-assisted diagnosis has been
testing of the resultant techniques on individuals, applied for detection of pathologic features in these
which is time-consuming and costly. Development patients. In one study, a GA was developed to detect
of musculoskeletal models enables computer the MS lesions of brain MRIs. The similarity index
simulation of movements to assess the effect of new of lesions determined by the GA and by a radiologist
modifications on the efficiency of training. Pei et was 87%.72
al,66 developed a robotic technique for physiotherapy The EEG is a useful diagnostic method to detect
of the lower limb. A GA was applied to generate the abnormal brain electrical discharges occurring
custom-made treatment plans for each patient. during a seizure. To design an automated system for
In another paper, a therapeutic robot was detection of abnormal EEG signals, several learning
designed for lower limb exercise. The system that algorithms (LM, Quickprop, Delta-bar delta, and
consisted of an ANN and a GA was capable of Momentum and Conjugate gradient) were used
learning the actions of a physiotherapist for each to train an ANN for EEG-based classification of
patient and mimicked its behavior in the absence of epileptic versus healthy individuals. A GA was used
a therapist.67 to find the optimal parameters for and architecture
of the ANN. The results demonstrated that the
Orthopedics LM method combined with the GA was the best
Biomedical engineering has offered great solutions algorithm for training the ANN, which reached a
to the field of orthopedic surgery. Total hip general success of 96.5% in its performance.73
arthroplasty (THA) has improved the management Several reports have suggested that mitochondrial
of various disabling hip joint diseases. Yet, failure dysfunction plays an important role in Parkinson’s
of the femoral stem of a THA can compromise disease. Since mitochondrial genetics has its
the success of treatment. Ishida et al,68 reported the idiosyncrasies, a simple comparison of mitochondrial
use of a GA in designing an optimized geometry of mutations between healthy and disease conditions
the femoral stem component. GAs have also been might not be so informative. Therefore, Smigrodzki
exploited to select the best design of tibial locking et al,74 devised a GA to detect biologically important
screws to reduce the probability of screw breakage patterns of mitochondrial mutations in Parkinson’s
 A li Gh aheri, et al. 413

patients. The proposed system was able to diagnose plasma protein binding of alkaloid drugs with an
Parkinson’s disease with 100% accuracy based on acceptable efficiency.
mutational patterns in mitochondrial DNA.
Health care management
Pharmacotherapy Proper management of monetary resources and
Pharmacovigilance, the study of safety and adverse personnel is an integral part of health systems all over
effects of drugs, is not only an integral part of currently- the world. One of the important elements of hospital
used drug assessment; it is also a crucial element in management which can improve patient servicing,
the evaluation of novel investigational medicines. satisfaction, and cost-effectiveness ratios is efficient
The clinical judgment of a pharmacotherapist to scheduling of patients admission. A mathematical
attribute an observed adverse effect to a drug is model was developed and optimized using a GA to
valuable yet implicit while algorithms can make a improve the patient scheduling in an ophthalmology
less arbitrary and more objective evaluation. Koh et hospital. The new algorithm was superior to the
al,75 developed a GA-based quantitative system for traditional “first come, first serve” model in that
the evaluation of adverse drug reactions. The new it shortened the waiting list, lowered the vacancy
scoring system was able to determine a probability rate of hospital beds, reduceed the preoperative
of the causality of an adverse drug reaction to a waiting time for patients, and increased the number
suspected drug with about 84% sensitivity and 71% of patients discharged from the hospital.78 Another
specificity. report showed that a combination of GA and
Tacrolimus is an immunosuppressive agent used particle swarm optimization, another powerful
to prevent rejection after organ transplantation. The metaheuristic algorithm, was able to improve patient
drug has highly variable pharmacokinetics and a scheduling, reduce time wastage, and increase patient
narrow therapeutic window making its blood level satisfaction.79
control an essential and difficult task. In an attempt In clinical laboratories, regular rotation of staff
to predict the blood concentration of tacrolimus based on their skills through different facilities
in liver-transplanted patients, an ANN algorithm is fundamental for maintaining job skills and
was developed. A GA was used to choose the best competence. GAs have been applied to improve
set of clinically significant candidate variables. For staff rotation scheduling in a clinical laboratory. In
validation, predicted results were compared to one report, the GA-based software was capable of
observed figures. The ANN was able to predict the planning the rotation of staff effectively, ensuring
blood level of tacrolimus, with 84% of data sets being maintenance of techniques and skills, saving time and
within a clinically acceptable range of 3 ng/ml of the the cost necessary for the scheduling process, and it
observed data.76 was associated with the satisfaction of responsible
Studies have shown that poor pharmacokinetics supervisory personnel.80
and lack of efficiency account for more than 50%
of failures in the process of drug development.
The traditional assessment of the efficacy and C O N C LU S I O N
pharmacokinetics of novel investigational agents In this paper, we introduced GAs and some of
in animal models is a costly and time-consuming their applications in various fields of medicine.
process. Therefore, computational methods Although GAs and some other metaheuristics are
have evolved to generate quantitative structure- inspired by biology, the experts of other fields of
pharmacokinetic relationship (QSPKR) models for science are more aware of them and these methods
rapid in silico screening of novel potential drugs. are frequently used to solve complex problems.
Zandkarimi et al,77 applied a GA to select the Due to the inherent complexity of medicine,
most suitable characteristics out of more than optimization methods could be of great value for
1480 descriptors of alkaloid drugs. These sets of physicians and medical researchers. The lack of an
characteristics were then extracted from known efficient interaction between computer scientists and
drugs for training an ANN to generate QSPKR physicians on the one hand and the unfamiliarity
prediction models. The new system was able to of complex mathematical formulas among the
predict the volume of distribution, clearance, and medical professions on the other is responsible for

O M A N M E D J, V O L 3 0 , N O 6 , N O V E M B E R 2015
414 A li Gh aheri, et al.

the interpretation of screening mammograms by US

this situation. Therefore, improving the interaction radiologists. Findings from a national sample. Arch Intern
and understanding between physicians, computer Med 1996 Jan;156(2):209-213.
scientists, and engineers, which could happen via 18. Elmore JG, Wells CK, Lee CH, Howard DH, Feinstein AR.
Variability in radiologists’ interpretations of mammograms.
joint journal clubs or attendance of physicians N Engl J Med 1994 Dec;331(22):1493-1499.
ground rounds and case report presentations, 19. Karnan M, Thangavel K. Automatic detection of the breast
border and nipple position on digital mammograms using
could solve the problem. Besides, improvement of genetic algorithm for asymmetry approach to detection of
interdisciplinary courses and efficient involvement of microcalcifications. Comput Methods Programs Biomed
2007 Jul;87(1):12-20.
engineering researchers in health care environments 20. Bushberg JT, Boone JM. Image Quality. The essential
and hospitals could offer new solutions for medical physics of medical imaging. second ed: Lippincott Williams
& Wilkins; 2011.
problems and new ideas for non-medical researchers. 21. Pereira DC, Ramos RP, do Nascimento MZ. Segmentation
and detection of breast cancer in mammograms combining
wavelet analysis and genetic algorithm. Comput Methods
Disclosure Programs Biomed 2014 Apr;114(1):88-101.
The authors declared no conflicts of interest. No funding was 22. Jiang J, Yao B, Wason AM. A genetic algorithm design
received for this study. for microcalcification detection and classification in
digital mammograms. Comput Med Imaging Graph 2007
Jan;31(1):49-61.
r ef er enc e s 23. Bevilacqua A, Campanini R, Lanconelli N. A distributed
1. Osman IH, Kelly JP. Meta-heuristics: an overview. Meta- genetic algorithm for parameters optimization to detect
Heuristics: Springer; 1996. p. 1-21. micro calcifications in digital mammograms. Applications
of Evolutionary Computing. Proceedings 2001;2037:278-
2. Dorigo M, Stützle T. The ant colony optimization 287.
metaheuristic: Algorithms, applications, and advances.
Handbook of metaheuristics: Springer; 2003. p. 250-285. 24. Yao B, Jiang JM, Peng YH. A CBR driven genetic algorithm
for microcalcification cluster detection. Engineering
3. Karaboga D, Basturk B. A powerful and efficient algorithm Knowledge in the Age of the Semantic Web. Proceedings
for numerical function optimization: artificial bee colony 2004;3257:494-496.
(ABC) algorithm. J Glob Optim 2007;39:459-471.
25. Wu WJ, Lin SW, Moon WK. Combining support vector
4. Mirjalili S, Mirjalili SM, Lewis A. Grey wolf optimizer. Adv machine with genetic algorithm to classify ultrasound
Eng Softw 2014;69:46-61. breast tumor images. Comput Med Imaging Graph 2012
5. Potvin J-Y, Smith KA. Artificial Neural Networks for Dec;36(8):627-633.
Combinatorial Optimization. Handbook of metaheuristics: 26. Vinterbo S, Ohno-Machado L. A genetic algorithm to select
Springer; 2003. p. 429-455. variables in logistic regression: example in the domain of
6. Henderson D, Jacobson SH, Johnson AW. The theory myocardial infarction. Proc AMIA Symp 1999;984-988.
and practice of simulated annealing. Handbook of 27. Sahiner B, Chan HP, Wei D, Petrick N, Helvie MA, Adler
metaheuristics: Springer; 2003. p. 287-319. DD, et al. Image feature selection by a genetic algorithm:
7. Rabanal P, Rodríguez I, Rubio F. Using river formation application to classification of mass and normal breast
dynamics to design heuristic algorithms. Unconventional tissue. Med Phys 1996 Oct;23(10):1671-1684.
Computation: Springer; 2007. p. 163-177. 28. de Carvalho Filho AO, de Sampaio WB, Silva AC, de
8. De Castro LN, Timmis J. Artificial immune systems: a new Paiva AC, Nunes RA, Gattass M. Automatic detection of
computational intelligence approach: Springer; 2002. solitary lung nodules using quality threshold clustering,
genetic algorithm and diversity index. Artif Intell Med 2014
9. Reeves C. Genetic algorithms. Handbook of metaheuristics: Mar;60(3):165-177.
Springer; 2003. p. 55-82.
29. Oliveira FP, Tavares JM. Medical image registration:
10. Hasancebi O, Erbatur F. Evaluation of crossover techniques a review. Comput Methods Biomech Biomed Engin
in genetic algorithm based optimum structural design. 2014;17(2):73-93.
Comput Struc 2000;78:435-448.
30. Valsecchi A, Damas S, Santamaria J, eds. An image
11. MathWorks. Genetic Algorithm. Available from: http:// registration approach using genetic algorithms. Evolutionary
www.mathworks.in/help/gads/what-is-the-genetic- Computation (CEC), 2012 IEEE Congress on; 2012:
algorithm.html. IEEE.
12. Krishnakumar K, Goldberg DE. Control-System 31. Fan Y, Jiang TZ, Evans DJ. Medical image registration using
Optimization Using Genetic Algorithms. J Guid Control parallel genetic algorithms. Applications of Evolutionary
Dyn 1992;15:735-740. Computing. Proceedings 2002;2279:304-314.
13. Shiraishi J, Li Q, Appelbaum D, Doi K. Computer-aided 32. Baum KG, Schmidt E, Rafferty K, Krol A, Helguera M.
diagnosis and artificial intelligence in clinical imaging. Evaluation of novel genetic algorithm generated schemes
Semin Nucl Med 2011 Nov;41(6):449-462. for positron emission tomography (PET)/magnetic
14. Gudmundsson M, El-Kwae EA, Kabuka MR. Edge resonance imaging (MRI) image fusion. J Digit Imaging
detection in medical images using a genetic algorithm. IEEE 2011 Dec;24(6):1031-1043.
Trans Med Imaging 1998 Jun;17(3):469-474. 33. Zhou J, Krishnan S, Chong V, Huang J, eds. Extraction of
15. Bhandarkar SM, Zhang YQ, Potter WD. An Edge- tongue carcinoma using genetic algorithm-induced fuzzy
Detection Technique Using Genetic Algorithm-Based clustering and artificial neural network from MR images.
Optimization. Pattern Recognit 1994;27:1159-1180. Engineering in Medicine and Biology Society, 2004
16. Karnan M, Thangavel K. Automatic detection of the breast IEMBS’04 26th Annual International Conference of the
border and nipple position on digital mammograms using IEEE; 2004: IEEE.
genetic algorithm for asymmetry approach to detection of 34. Duraipandian S, Zheng W, Ng J, Low JJ, Ilancheran A,
microcalcifications. Comput Methods Programs Biomed Huang Z. In vivo diagnosis of cervical precancer using
2007 Jul;87(1):12-20. Raman spectroscopy and genetic algorithm techniques.
17. Beam CA, Layde PM, Sullivan DC. Variability in Analyst 2011 Oct;136(20):4328-4336.
 A li Gh aheri, et al. 415

35. Wold S, Sjostrom M, Eriksson L. PLS-regression: a basic tool Comparison of a genetic algorithm neural network with
of chemometrics. Chemom Intell Lab Syst 2001;58:109- logistic regression for predicting outcome after surgery for
130. patients with nonsmall cell lung carcinoma. Cancer 1997
36. Ooi CH, Tan P. Genetic algorithms applied to multi- Apr;79(7):1338-1342.
class prediction for the analysis of gene expression data. 55. Lee J, Govindan S, Celi LA, Khabbaz KR, Subramaniam B.
Bioinformatics 2003 Jan;19(1):37-44. Customized Prediction of Short Length of Stay Following
37. Dolled-Filhart M, Rydén L, Cregger M, Jirström K, Elective Cardiac Surgery in Elderly Patients Using a Genetic
Harigopal M, Camp RL, et al. Classification of breast Algorithm. World J Cardiovasc Surg 2013 Sep;3(5):163-
cancer using genetic algorithms and tissue microarrays. Clin 170.
Cancer Res 2006 Nov;12(21):6459-6468. 56. Güler I, Polat H, Ergün U. Combining neural network and
38. Tan C, Chen H, Wu T, Xia C. Modeling the relationship genetic algorithm for prediction of lung sounds. J Med Syst
between cervical cancer mortality and trace elements based 2005 Jun;29(3):217-231.
on genetic algorithm-partial least squares and support vector 57. Engoren M, Plewa M, O’Hara D, Kline JA. Evaluation of
machines. Biol Trace Elem Res 2011 Apr;140(1):24-34. capnography using a genetic algorithm to predict PaCO2.
39. van Gestel T, Suykens JA, Baesens B, et al. Benchmarking Chest 2005 Feb;127(2):579-584.
least squares support vector machine classifiers. Mach Learn 58. Elveren E, Yumuşak N. Tuberculosis disease diagnosis using
2004;54:5-32. artificial neural network trained with genetic algorithm. J
40. Bozcuk H, Bilge U, Koyuncu E, Gulkesen H. An application Med Syst 2011 Jun;35(3):329-332.
of a genetic algorithm in conjunction with other data mining 59. Castiglione F, Poccia F, D’Offizi G, Bernaschi M.
methods for estimating outcome after hospitalization in Mutation, fitness, viral diversity, and predictive markers
cancer patients. Med Sci Monit 2004 Jun;10(6):CR246- of disease progression in a computational model of HIV
CR251. type 1 infection. AIDS Res Hum Retroviruses 2004
41. Khalil AS, Bouma BE, Kaazempur Mofrad MR. A combined Dec;20(12):1314-1323.
FEM/genetic algorithm for vascular soft tissue elasticity 60. Bernaschi M, Castiglione F. Design and implementation
estimation. Cardiovasc Eng 2006 Sep;6(3):93-102. of an immune system simulator. Comput Biol Med 2001
42. Zhou X, Wang H, Wang J, Wang Y, Hoehn G, Azok J, et Sep;31(5):303-331.
al. Identification of biomarkers for risk stratification of 61. Castiglione F, Pappalardo F, Bernaschi M, Motta S.
cardiovascular events using genetic algorithm with recursive Optimization of HAART with genetic algorithms and
local floating search. Proteomics 2009 Apr;9(8):2286-2294. agent-based models of HIV infection. Bioinformatics 2007
43. Tu C, Zeng Y, Yang X. A new approach to detect QRS Dec;23(24):3350-3355.
complexes based on a histogram and genetic algorithm. J 62. Nazareth DP, Brunner S, Jones MD, Malhotra HK,
Med Eng Technol 2005 Jul-Aug;29(4):176-180. Bakhtiari M. Optimization of beam angles for intensity
44. Nguyen LB, Nguyen AV, Ling SH, Nguyen HT. Combining modulated radiation therapy treatment planning using
genetic algorithm and Levenberg-Marquardt algorithm genetic algorithm on a distributed computing platform. J
in training neural network for hypoglycemia detection Med Phys 2009 Jul;34(3):129-132.
using EEG signals. Conf Proc IEEE Eng Med Biol Soc. 63. Ezzell GA, Gaspar L. Application of a genetic algorithm to
2013;2013:5386-5389. optimizing radiation therapy treatment plans for pancreatic
45. Ling SS, Nguyen HT. Genetic-algorithm-based multiple carcinoma. Med Dosim 2000;25(2):93-97.
regression with fuzzy inference system for detection of 64. Wu X, Zhu Y. A mixed-encoding genetic algorithm with
nocturnal hypoglycemic episodes. IEEE Trans Inf Technol beam constraint for conformal radiotherapy treatment
Biomed 2011 Mar;15(2):308-315. planning. Med Phys 2000 Nov;27(11):2508-2516.
46. Hoh JK, Cha KJ, Park MI, Ting Lee ML, Park YS. 65. Yu Y, Schell MC, Zhang JB. Decision theoretic steering
Estimating time to full uterine cervical dilation using genetic and genetic algorithm optimization: application to
algorithm. Kaohsiung J Med Sci 2012 Aug;28(8):423-428. stereotactic radiosurgery treatment planning. Med Phys
47. Marinakis Y, Dounias G, Jantzen J. Pap smear diagnosis using 1997 Nov;24(11):1742-1750.
a hybrid intelligent scheme focusing on genetic algorithm 66. Pei Y, Kim Y, Obinata G, Hase K, Stefanov D. Trajectory
based feature selection and nearest neighbor classification. planning of a robot for lower limb rehabilitation. Conf Proc
Comput Biol Med 2009 Jan;39(1):69-78. IEEE Eng Med Biol Soc. 2011;2011:1259-1263.
48. Boisvert MR, Koski KG, Burns DH, Skinner CD. Early 67. Aminiazar W, Najafi F, Nekoui MA. Optimized intelligent
prediction of macrosomia based on an analysis of second control of a 2-degree of freedom robot for rehabilitation of
trimester amniotic fluid by capillary electrophoresis. lower limbs using neural network and genetic algorithm. J
Biomark Med 2012 Oct;6(5):655-662. Neuroeng Rehabil 2013;10:96.
49. Yu J, Wang Y, Chen P. Fetal weight estimation using the 68. Ishida T, Nishimura I, Tanino H, Higa M, Ito H, Mitamura
evolutionary fuzzy support vector regression for low-birth- Y. Use of a genetic algorithm for multiobjective design
weight fetuses. IEEE Trans Inf Technol Biomed 2009 optimization of the femoral stem of a cemented total hip
Jan;13(1):57-66. arthroplasty. Artif Organs 2011 Apr;35(4):404-410.
50. Ocak H. A medical decision support system based on 69. Hsu CC, Chao CK, Wang JL, Lin J. Multiobjective
support vector machines and the genetic algorithm optimization of tibial locking screw design using a genetic
for the evaluation of fetal well-being. J Med Syst 2013 algorithm: Evaluation of mechanical performance. J Orthop
Apr;37(2):9913. Res 2006 May;24(5):908-916.
51. Latkowski T, Osowski S. Computerized system for 70. Amaritsakul Y, Chao CK, Lin J. Multiobjective
recognition of autism on the basis of gene expression optimization design of spinal pedicle screws using neural
microarray data. Comput Biol Med 2015 Jan;56:82-88. networks and genetic algorithm: mathematical models
52. Lin TC, Liu RS, Chao YT, Chen SY. Classifying subtypes and mechanical validation. Comput Math Methods Med
of acute lymphoblastic leukemia using silhouette statistics 2013;2013:462875.
and genetic algorithms. Gene 2013 Apr;518(1):159-163. 71. Jaremko JL, Poncet P, Ronsky J, Harder J, Dansereau
53. Wang TH, Chang YL, Peng HH, Wang ST, Lu HW, J, Labelle H, et al. Genetic algorithm-neural network
Teng SH, et al. Rapid detection of fetal aneuploidy using estimation of cobb angle from torso asymmetry in scoliosis.
proteomics approaches on amniotic fluid supernatant. J Biomech Eng 2002 Oct;124(5):496-503.
Prenat Diagn 2005 Jul;25(7):559-566. 72. Khotanlou H, Afrasiabi M. Feature Selection in Order to
54. Jefferson MF, Pendleton N, Lucas SB, Horan MA. Extract Multiple Sclerosis Lesions Automatically in 3D
Brain Magnetic Resonance Images Using Combination of

O M A N M E D J, V O L 3 0 , N O 6 , N O V E M B E R 2015
416 A li Gh aheri, et al.

Support Vector Machine and Genetic Algorithm. J Med patients. Ther Drug Monit 1999 Feb;21(1):50-56.
Signals Sens 2012 Oct;2(4):211-218. 77. Zandkarimi M, Shafiei M, Hadizadeh F, Darbandi MA,
73. Koçer S, Canal MR. Classifying epilepsy diseases using Tabrizian K. Prediction of pharmacokinetic parameters
artificial neural networks and genetic algorithm. J Med Syst using a genetic algorithm combined with an artificial neural
2011 Aug;35(4):489-498. network for a series of alkaloid drugs. Sci Pharm 2014
74. Smigrodzki R, Goertzel B, Pennachin C, Coelho L, Mar;82(1):53-70.
Prosdocimi F, Parker WD Jr. Genetic algorithm for analysis 78. Zhang LM, Chang HY, Xu RT. The Patient Admission
of mutations in Parkinson’s disease. Artif Intell Med 2005 Scheduling of an Ophthalmic Hospital Using Genetic
Nov;35(3):227-241. Algorithm. Adv Mat Res 2013;756:1423-1432.
75. Koh Y, Yap CW, Li SC. A quantitative approach of using 79. Du G, Jiang Z, Yao Y, Diao X. Clinical pathways
genetic algorithm in designing a probability scoring system scheduling using hybrid genetic algorithm. J Med Syst 2013
of an adverse drug reaction assessment system. Int J Med Jun;37(3):9945.
Inform 2008 Jun;77(6):421-430. 80. Boyd JC, Savory J. Genetic algorithm for scheduling of
76. Chen HY, Chen TC, Min DI, Fischer GW, Wu YM. laboratory personnel. Clin Chem 2001 Jan;47(1):118-123.
Prediction of tacrolimus blood levels by using the neural
network with genetic algorithm in liver transplantation