Hybrid bio-inspired computing in medical image data analysis: A review

Abstract

Inspired by the fundamentals of biological evolution, bio-inspired algorithms are becoming increasingly popular for developing robust optimization techniques. These metaheuristic algorithms, unlike gradient descent methods, are computationally more efficient and excel in handling higher order multi-dimensional and non-linear.

OBJECTIVES

To understand the hybrid Bio-inspired algorithms in the domain of Medical Imaging and its challenges of hybrid bio-inspired feature selection techniques.

METHOD

The primary research was conducted using the three major indexing database of Scopus, Web of Science and Google Scholar.

RESULT

The primary research included 198 articles, after removing the 103 duplicates, 95 articles remained as per the criteria. Finally 41 articles were selected for the study.

CONCLUSION

We recommend that further research in the area of bio-inspired algorithms based feature selection in the field of diagnostic imaging and clustering. Additionally, there is a need to further investigate the use of Deep Learning hybrid models integrating the bio-inspired algorithms to include the strengths of each models that enhances the overall hybrid model.

Keywords

Bio-inspired optimization feature selection metaheuristics literature review hybrid algorithms

1. .Introduction

Bio-inspired algorithms are computational methodologies that mimic natural systems, behavior such as evolution, neural networks and swarm intelligence. Genetic Algorithms (GAs) [1] are built on what is called the principles of natural selection and genetics mimicking the evolutionary process where the most fit individuals survive and pass their traits to future generations. The name Swarm Intelligence (SI) [2] is derived from various actions by ants and bees that are social animals with the aim of using a multiple simple agents to achieve an objective similar to natural colonies. The best outcome will be collectively found by particles moving through a solution space in Particle Swarm Optimization (PSO) [3] just like the flocking behavior of birds. Alternatively, Ant Colony Optimization (ACO) [4] is founded on the foraging behavior of ants in which synthetic ants explore search spaces by following pheromone trails to find best paths. Ultimately, Bacterial Foraging Optimization (BFO) [5] algorithms are based on bacteria feeding and they employ virtual bacteria to explore solution spaces and identify optimal solutions in complex environments using bacterial foraging strategies.

1.1. .Scope of discussion

The main objective of this Review is to examine and validate the use and effectiveness of several hybrid bio-inspired algorithms in the field of Medical Image Data Analysis.

1.2. .Our contribution

This review article aims to examine the hybrid-bio inspired algorithms in the domain of Medical Image Analysis. The goal is to contribute to future research and find research gaps in classification problems in medical imaging and diagnosis. Figure 1 shows the model diagram of conventional hybrid approaches of Hybrid Bio-inspired Algorithms.

Figure 1.

Model diagram for hybrid bio-inspired algorithm.

2. .Literature review

The article [6] proposed a hybrid expert system (HES) that are designed to handle the complexities in the area of Artificial Intelligence particularly in medical imaging and diagnosis. The HES utilizes a combination of cognitive computing approaches, including artificial neural networks (ANN), fuzzy logic, and evolutionary computation, to address the issues associated with knowledge acquisition, human reasoning and providing explanations to the results obtained by the neural network. The algorithm proposed utilized the neuro-fuzzy methodology in along with the genetic algorithm in order to enhance the overall architecture of the neural network and generate rules which are explainable and understandable. The proposed model was applied on the dataset of epilepsy where the classification accuracy reached between 63.6%–83.3%, displaying the practical applicability to manage the complex dataset as well as providing reliable diagnosis. The findings underscore the capacity of the system to improve medical decision-making through the provision of lucid, comprehensible explanations and the optimization of neural network typologies, rendering it a valuable instrument for healthcare practitioners.

The Crow Search Algorithm (CSA) [7] is a bio-inspired optimization approach modeled after the foraging behavior of crows. Despite its conceptual appeal, CSA typically confronts problems such as low converging rates and being held in local optima. To address these restrictions, the Chaotic Crow Search Algorithm (CCSA) was proposed as a revolutionary meta-heuristic optimizer. CCSA incorporates 10 unique chaotic maps to boost both the performance and convergence speed of CSA. In an empirical study incorporating feature selection tasks across 20 benchmark datasets, CCSA displays higher performance relative to established optimization techniques. Notably, the Sine chaotic map appears as very beneficial in boosting the efficacy of CCSA. Moving forward, prospective routes for further research include the application of CCSA to more sophisticated scientific and technical challenges, as well as the exploration of other chaotic maps. The Algorithm 1 represents the Chaotic Crow Search Algorithm (CCSA) approach.

The research [8] introduces a hybrid optimization method, combining the Grey Wolf Optimizer (GWO) and the Sine Cosine method (SCA), to boost the performance of solving optimization problems. The hybrid technique, termed HGWOSCA, harnesses the exploration capabilities of SCA and the exploitation efficiency of GWO. The algorithm is tested on 22 benchmark functions, five biological datasets, and one sine dataset. Results reveal that HGWOSCA outperforms other meta-heuristic algorithms, such as Particle Swarm Optimization (PSO), Ant Lion Optimizer (ALO), and Whale Optimization Algorithm (WOA), in terms of discovering the global optimal solutions and convergence speed. Numerical and statistical studies reveal that HGWOSCA provides greater performance, providing improved accuracy and stability compared to solo GWO and SCA. The study concludes that the suggested hybrid algorithm effectively addresses complicated optimization problems, exhibiting considerable gains in solution quality and computational efficiency, the Algorithm 2 depicts the Hybrid GWO-SCA Algorithm.

The authors [9] underline the crucial role played by computational intelligence in developing medical diagnosis and therapy through the integration of mathematical simulation and programming. The core focus of the study revolves around the simulation of medical tests for pulmonary disorders, applying bio-inspired approaches specifically developed to autonomously identify sick tissues in input x-ray pictures. The suggested approach combines specialized fitness conditions optimized for heuristic algorithms, enabling the detection of specific traits associated with pulmonary diseases such as pneumonia, sarcoidosis, and cancer. The system presents detection results created by numerous algorithms, allowing doctors a full variety of specialist-like insights. Benchmark studies undertaken affirm the effectiveness and potential of this technology in detecting lung illnesses within x-ray pictures acquired from varied clinical contexts.

In [10] Authors proposed a bio-inspired optimization technique with practical applications, has often been overlooked due to its perceived excessive randomness, which can result in inaccurate outcomes. This research endeavors to overcome this limitation by proposing modified GA approaches. These modifications primarily focus on developing reproduction operators to mitigate the randomness inherent in conventional GA. The efficacy of these proposed methods is demonstrated in the domain of medical image classification, where they achieve an impressive accuracy rate of 98% compared to alternative techniques. Furthermore, three distinct GA approaches are introduced for feature selection, resulting in approximately 4% to 6% improvement in accuracy while simultaneously reducing the number of features required. Looking ahead, future avenues of exploration include experimenting with various optimization methodologies, further refining conventional GA through additional modifications, exploring alternative feature sets, and integrating neural classifiers to enhance system efficiency.

In [11], Semi-supervised learning leverages unlabeled data to augment supervised learning in scenarios where labeled data is limited. Active learning, on the other hand, complements semi-supervised learning by engaging domain experts to selectively label instances. However, prevailing active learning methods often incur high computational costs and may include redundant instances, thus compromising performance of the model and diagnosis. To mitigate this, a hybrid system merging active learning with PSO algorithms was developed to streamline labeling expenses and enhance classifier efficiency as shown in Algorithm 3. This hybrid system integrated a new uncertainty metric into the PSO algorithm to identify the most informative instances from vast pools of unlabeled medical data. Results demonstrate substantial performance increases over typical uncertainty algorithms, with similar efficacy to fully supervised and semi-supervised systems while needing reduced labeling effort cmaped to original method. For further researc initiatives could go into combining alternate evolutionary optimization techniques and studying new forms of the particle swarm optimization process, while evaluating various classification frameworks.

The main objective of this work [12] is to build efficient Machine Learning (ML) classification models suited for healthcare data harnessing Bio-Inspired Optimization (BIO) methods. Particularly, the paper addresses the difficulty provided by high-dimensional data by using two BIO algorithms: Genetic Algorithm (GA) and Binary Particle Swarm Optimization (BPSO) as shown in Algorithm 4. The dataset utilized includes 756 observations collected from 755 variables belonging to 252 people who were diagnosed with Parkinson’s Disease (PD). To ensure impartial results, the MaxAbsolute parameter scaling method is employed for data normalization, alongside applying a one-hold cross-validation methodology.

The study analyzes 11 ML classifiers, such as Logistic Regression, Support Vector Machines, Neural Networks etc., to discover the ideal selection of features leading to the maximum accuracy in classification. Results demonstrate that GA-inspired MLP achieved a significant dimensionality reduction of 52.32% and an accuracy rate of 85.1%. Additionally, GA-inspired AB yielded the maximum classification accuracy of 90.7%, coupled by a dimensionality reduction of 41.43%. Notably, BPSO-inspired GNB achieved a dimensionality reduction of 47.14% and an accuracy rate of 79.3%. Moreover, the BPSO-MLP model got the maximum classification accuracy of 89%, alongside a dimensionality reduction of 46.48%. These findings underline the usefulness of BIO algorithms in boosting classification accuracy while simultaneously reducing dimensionality, thereby having substantial promise for healthcare data analysis.

Fuzzy c-means (FCM) is commonly applied in unsupervised clustering throughout numerous applications. However, when faced with complicated problems like medical picture data, FCM typically becomes stuck in local minima, resulting in suboptimal clustering conclusions. To address this limitation, Authors [13] proposed PSO, a metaheuristic optimization algorithm known for its global search capability, is introduced. A novel hybrid approach, combining FCM with PSO, is proposed to mitigate FCM’s issues and enhance clustering results as shown in Algorithm 5. The efficacy of this hybrid algorithm is evaluated using both triangular synthetic datasets and publicly available datasets. Results indicate significant improvements over other algorithms in terms of average similarity measure, as determined by Friedman’s statistical test. Specifically, the proposed hybrid system exhibits 33% much better efficiency on triangular dataset and upto 30% better efficiency on the real-brain image datasets compared to alternative algorithms. These findings underscore the effectiveness of the hybrid approach in optimizing clustering outcomes, particularly in challenging scenarios like medical image analysis.

As per the approach [14] integrates a novel feature selection technique alongside an enhanced cuckoo search optimization algorithm (CS) that incorporates fractional-order calculus (FO) and heavy-tailed distributions, aiming to enhance algorithmic performance. The classification task encompasses distinguishing among three classes: COVID-19 infected patients, pneumonia patients and normal patients. Various heavy-tailed distributions, including Cauchy, Mittag-Leffler, Weibull and Pareto, are employed in the proposed FO-CS variants. The effectiveness of the proposed methodology is assessed across eighteen UCI datasets, where it is benchmarked against established optimization algorithms. Results demonstrate superior performance in convergence behavior, particularly notable with the Weibull distribution. Notably, the FO-CS method utilizing the Weibull distribution emerges as particularly effective in classifying COVID-19 datasets as shown in Algorithm 6. Future endeavors will explore the broader applicability of the proposed approach across diverse domains.

Accurate segmentation of pre-treatment and post-treatment organs remains a formidable challenge in medical image analysis [15], particularly when dealing with limited datasets. In such scenarios, unsupervised models become imperative. This paper introduces a novel unsupervised clustering model termed DLPSO-NIFCM for organ segmentation. The model incorporates a primary convex objective function and applies algebraic transformations to enhance suitability. A dynamically learned PSO is employed to derive initial cluster centroids. Results demonstrate precise clustering with notably faster convergence times compared to existing state-of-the-art methods. The suggested DLPSO-NIFCM model appears as an intriguing option for clustering and CT image segmentation inside Computer-Aided Diagnosis (CAD) systems, bringing significant breakthroughs in medical image analysis and clinical diagnosis.

The study conducted by [16] offers a model aimed to classify COVID-19 infected individuals based on their Chest X-Ray (CXR) images. The model applies a Hybrid Social Group Optimization technique to successfully extract the features from the given CXR images as input to the model. Subsequently, the selected feature set is then passed through the classification employing several classifiers using Algorithm 7. Significantly, the suggested model obtains a remarkable accuracy of 99.65% while applying the Support Vector Classifier, surpassing the greatest accuracy of 99.27% attained by deep learning competitors. Notably, the suggested method demonstrates both greater precision and much lower duration of training compared to all currently available state-of-the-art deep learning methods. This study provides an innovative and effective technique to COVID-19 patient classification based on CXR pictures, revealing promising potential for improved treatment and diagnosis management in clinical settings.

Detecting brain tumors early offers a huge problem in healthcare diagnosis. To address this, a computer-assisted brain tumor identification approach is suggested in [17], using empirical wavelet transform-based local binary pattern variant features and ant-lion optimization. Initially, the input MRI image undergoes improvement via the DFHE algorithm. Subsequently, relevant modes are identified from the input brain MRI iamge using EWT, from which five unique local binary variations are produced. The ant-lion bio-inspired algorithm was applied in order to reduce the overall dimensionality of dataset by removing the duplicate descriptors and preserving the relevant features. In the final step the SVM (Support Vector Machine) a machne learning approach is applied to classify the input MRI image as healthy or tumorous. The results of the proposed model over publicly accessible datasets proves the robustness and efficacy of the proposed model for an early brain tumor detection.

The research [18] provides a hybrid bio-inspired algorithm for identifying of COVID-19 patients without the need of human expertise and intervention. The empirical data shows that the model achieves 99.34% accuracy displaying its resilience and usefulness in binary COVID-19 diagnosis.

The research [19] proposed a hybrid approach for COVID-19 diagnosis and highlights the need of higher order of valid diagnosis and examinations.Particularly, Chest X-ray imaging acts as a critical technique in diagnosing COVID-19 and related pneumonias. The study introduces a specialized framework tailored for COVID-19 detection in X-ray images. The research delves into various radiomic features, normalization methods, and feature ranking techniques to enhance detection accuracy. The proposed framework integrates a cutting-edge classifier with Gauss-map-based chaotic particle swarm optimization and neural networks to optimize performance. Experimental results showcase exceptional performance, with accuracy and ROC area metrics surpassing 99% for coronavirus classification. This research underscores the potential of advanced imaging techniques and computational methodologies in aiding timely and accurate diagnosis of COVID-19, thereby contributing to better patient care and COVID-19 management.

Table 1
BALO-S and BALO-V binary ant lion optimization.

Algorithm name Dataset Result accuracy

Hybrid genetic algorithm Nutritional anemia datasets 91.2%

BALO-S Brain MR images dataset 97.22%

BALO-V Brain MR images dataset 97.22%

Algorithm name	Dataset	Result accuracy
Hybrid genetic algorithm	Nutritional anemia datasets	91.2%
BALO-S	Brain MR images dataset	97.22%
BALO-V	Brain MR images dataset	97.22%

The research [20] delves into the comprehensive examination of baseline, laboratory, and CT features among two distinct groups of COVID-19 patients: those in early and critical stages. The aim is to propose a detection model based on disease severity manifestations. Various Machine Learning, Deep Learning and Hybrid Learning models are deployed to analyze CT scan images for feature extraction and prediction of CT scores. Predicted CT scores, along with clinical, laboratory, and CT image features, serve as inputs for training regression models to forecast the COVID Criticality (CC) Score. Statistical and uni-variate logistic regression analysis are applied to reduce the features associated with COVID-19. Results indicate that AlexNet+Lasso demonstrates superior performance in CT score prediction, while VGG-16+Linear Regression excels in predicting CC scores. Correlation analysis underscores the significance of incorporating other features in CC Score prediction, revealing strong correlation coefficients. This research facilitates the rapid classification of COVID patients into early or severe stages by radiologists and automates the prediction of COVID Criticality Score, potentially streamlining treatment procedures and enhancing patient care.

Multi-threshold image segmentation (MIS) is a widely recognized technique in image processing. Despite intelligent algorithm applications, existing methods often encounter challenges related to local optimal solutions. To address this, a novel approach named EHSSA was introduced [21], enhancing the Salp Swarm Algorithm (SSA) for MIS applications. EHSSA incorporates an efficient mechanism to enhance global search capabilities and mitigate local optima. Comparative research conducted at IEEE CEC2014 demonstrated the superior prowess of the proposed algorithm. EHSSA’s effectiveness was further validated through successful image segmentation tasks using datasets such as the Berkeley segmentation dataset and breast cancer microscopic images. The primary goal of medical image segmentation is to expedite the extraction of objects of interest, enhancing diagnostic accuracy, and aiding physicians in making informed decisions for patient rehabilitation.

In response to the pressing need for improved COVID-19 diagnostic methods, researchers have developed [22] an AI-based framework for highly accurate COVID-19 detection in CT lung images. However, the scarcity of publicly available CT datasets presented a significant obstacle. To address this challenge, they proposed an algorithm that leverages Convolutional Neural Networks (CNN), pre-trained models, and the Sparrow search algorithm (SSA) for COVID-19 classification. The research involved the utilization of various pre-trained CNN models, such as MobileNetV3Large and SeNet154, with hyperparameters optimized using SSA. Two datasets were employed, comprising COVID-19 lung CT scans and the Large COVID-19 CT scan slice dataset, totaling 14,486 and 17,104 images, respectively. The proposed framework demonstrated superior performance compared to other models, achieving an accuracy of 99.74% for two-class classification and 98% for three-class classification. This advancement holds promise for enhancing COVID-19 diagnosis accuracy and efficacy in clinical settings.

The COVID-19 pandemic has impacted people around the globle both mentally and physically and thier daily life, underscoring the urgent need for early detection and precise diagnosis to mitigate mortality and transmission rates. Chest X-ray (CXR) images have emerged as a valuable tool due to their robust lung features, affordability, and rapid interpretation by radiologists. The study [23] introduces CXGNet, a three-stage CXR-based COVID-19 classification model. Utilizing deep learning techniques coupled with an enhanced GWO and GA (EGWO-GA), CXGNet demonstrates superior performance. The model acheived the accuracy of 94.00%, 97.05% and 100% accuracy in 4-class, 3-class and binary class classification respectively thus surpassing conventional methods. These results highlight CXGNet as a promising tool for accurate COVID-19 diagnosis, offering potential benefits in clinical practice and public health management.

Alzheimer’s Disease (AD), marked as gradual decline in the cognitive skills eventually which results in dementia, presents a significant challenge in diagnosis and management. Current diagnostic practices heavily rely on symptom observation by family members, with MRI scans playing a pivotal role in disease assessment. Leveraging recent advancements in medical image analysis through machine learning, this study explores innovative avenues for efficient feature extraction from MRI data. Traditional methods such as Convolutional Neural Networks (CNNs) often generate extensive feature sets requiring substantial computational resources. In this context, our approach introduces the Optimized Crow Search Algorithm (OCSA) to facilitate early detection of AD directly from raw MRI data, producing a dense and highly indicative embedding. The methodology in [24] demonstrates remarkable efficacy, achieving an impressive accuracy rate of 98.62% in AD diagnosis by leveraging learned mappings between image embeddings and corresponding labels. This novel approach holds promise for enhancing early detection and management strategies for AD, potentially contributing to improved patient outcomes and quality of life.

The paper presents [25] an innovative approach that integrates image fusion, artificial life (AL), and a genetic algorithm (GA) for breast cancer ultrasound (US) images. The method employs memory-enabled artificial agents (TAs) within fusion images to discern edge patterns and delineate tumors. Comparative evaluation against 16 existing segmentation methods on intricate breast cancer images showcases the efficiency and robustness of the proposed model, underscoring its efficacy in addressing challenging scenarios.

The Coronavirus Optimization Algorithm (COVIDOA) is presented as an innovative optimization framework in this study [26], drawing inspiration from SARS-CoV-2 and employing evolutionary search strategies. Comprehensive evaluations across various problem sets, including 20 classical benchmark challenges, five CEC benchmark test functions, and five CEC 2011 real-world scenarios, demonstrate COVIDOA’s efficacy. Notably, COVIDOA consistently outperforms existing optimization algorithms across a majority of test instances, while maintaining competitive performance in remaining cases. Characterized by robust exploitation and exploration capabilities, as well as accelerated convergence rates, COVIDOA exhibits clear advantages over conventional metaheuristics. The findings collectively highlight COVIDOA’s superior efficacy relative to established optimization methodologies.

Glaucoma, a prominent cause of vision impairment, underscores the importance of expert ophthalmologist evaluation, traditionally conducted through manual eye screening or fundus image analysis. However, these methods are labor-intensive, reliant on human expertise, and prone to intra-observer variability. To mitigate these challenges, the medical imaging domain explores AI-based CAD systems for glaucoma detection in [27], merging ML for classification with bio-inspired computing for feature selection. This work provides two unique two-layered techniques (BA-BCS, BCS-PSO) predicated on PSO, BCS, and Bat Algorithm (BA), along with individual assessments the bio-inspired algorithms. Implementation of these approaches on benchmark datasets (ORIGA and REFUGE) produces a maximum accuracy of 98.95%, providing valuable answers for ophthalmologists and investigators while improving the area of glaucoma diagnostics.

Breast cancer, a widespread and life-threatening ailment impacting women globally, underscores the urgency of early detection for appropriate treatment and halting disease progression. The study [28] introduces a novel patient detection strategy aimed at identifying breast cancer patients at an earlier stage. The strategy comprises two integral components: data pre-processing and patient detection. The proposed method, named NHFSM, presents a hybrid feature selection approach that amalgamates information gain, bat algorithm, and particle swarm optimization. NHFSM surpasses existing feature selection methodologies, demonstrating superior performance in accuracy, precision, sensitivity/recall, and F-measure. Notably, NHFSM achieves a low error rate of 0.03, positioning it as a promising and accurate tool for breast cancer diagnosis with an impressive accuracy rate of 97%.

Differential Evolution (DE) stands as a widely utilized metaheuristic algorithm which is used under several optimization. Nonetheless, the DE algorithm is prone to stagnation and premature convergence, particularly when confronted with complex problem landscapes. To address these shortcomings, this study introduces Enhanced DE (EDE), a novel algorithm engineered to surmount such challenges. EDE integrates two distinct strategies: an Artificial Bee Colony (ABC)-inspired Shrink-wrap (SW) mechanism for augmented global exploration capability and an Elite Levy Spreading (ELS) strategy dedicated to enhancing the quality of optimal solutions in [29]. Through scalability experiments, EDE’s performance is systematically evaluated and compared against SOTA and champion algorithms using rigorous statistical analyses. Furthermore, EDE is validated through image segmentation experiments conducted on nine breast cancer images, showcasing its satisfactory performance and potential utility in practical applications.

This research evaluated the critical need for early oral cancer diagnosis to prevent fatalities. It introduces an intelligent method as a supplementary tool to reduce human errors in the complex early-stage diagnosis. The approach involves Reinforcement Learning-based image segmentation, Gabor wavelet-based feature extraction, and classification using an RBF-kernel-based SVM. A Modified Locust Swarm Optimization (MLSO) algorithm is applied for feature selection and SVM configuration optimization, reducing system complexity in [30]. Testing on the “Oral Cancer images” dataset showed outstanding performance, with a 96.94% accuracy rate, highlighting its efficiency when compared to other methodologies. The proposed technique also achieves 93.89% sensitivity, 92.37% specificity, 92.37% PPV, and 96.94% NPV.

Brain tumors are associated with high mortality rates, emphasizing the importance of early detection for enhanced patient outcomes. Magnetic resonance imaging (MRI) serves as a cornerstone in tumor assessment, yet manual analysis methods are susceptible to errors due to the extensive volume of data involved. To mitigate this challenge, a novel deep learning-based classification framework is proposed in [31], comprising five distinct modules. The framework initiates with skull-stripping and guided bilateral filtering of MRI images, followed by precise tumor region delineation utilizing a thresholding approach. Subsequently, prominent texture and edge features are extracted via an optimized Gabor wavelet transform methodology. Optimal feature subset selection is conducted employing the black widow adaptive red deer optimization algorithm. Finally, the chosen features are fed into a hybrid Elman bidirectional LSTM network for robust classification. The framework achieves an exceptional accuracy of 98.4%, significantly mitigating misclassifications and facilitating early diagnosis and treatment stratification for brain tumors.

The research in [32] proposed a hybrid algorithm of combining the Whale Optimization Algorithm (WOA) and Adaptive PSO (APSO) algorithm with Convolutional Neural Networks for detecting lung cancer from CT images. The proposed hybrid model of WOA-APSO with CNNs achieved the accuracy of 97.18% outperforming traditional algorithms or CNNs with much efficient feature selection and optimized computational efficiency. Though the study only included 120 CT images but shows the intrinsic property of bio-inspired algorithms to produce good classification accuracy over small datasets. However further research is required to validate the same over the large datasets as CNNs works well over large datasets to ensure the robustness and applicability for medical image diagnosis.

The enhanced Whale Optimization Algorithm (eWOA) have shown its applicability and improvements as shown in [33] outcomes. The model converges better and produce better accuracy and precision than its counterparts. Methodologies such as Lévy flights, chaotic maps, opposition-based learning, and adaptive parameter tactics have been particularly effective. Though due to its limitations more practical case studies and application on medical imaging require further investigations.

The paper [34] proposed a novel algorithm for glaucoma diagnosis using optimized KELM classifier which is a hybrid of correlation-based feature selection algorithm integrated with bio-inspired algorithms. The three main bio-inspired algorithms that were understudied this research were Lion Optimization (LOA), Grey Wolf Optimization (GWO), and Salp Swarm Optimization (SSA). These bio-inspired algorithm selected the feature considering their correlation hence enhancing its overall performance. The results shows the accuracy to be 99.61% and with 5-fold cross-validation the accuracy was 98.78%.

In the paper [35] proposed a hybrid algorithm with combining BFOA and EPOA for feature selection. The proposed novel algorithm was tested on different dimensional datasets and iterations using 10-fold and 5-fold cross validation. The hybrid algorithm of BFOA and EPOA combine the strength of each other (global search and local search respectively) demonstrated superior performance over various metrics hence optimizing the overall feature selection process defeating individual algorithm providing the robust solution for diagnosis.

The study [36] proposed a hybrid algorithm using the Deep Belief Network (DBN) which was optimized using the Cuckoo Search Algorithm (CSA). The model applied the hamming distance for feature selection purpose, this enhances the overall accuracy of the DBN as the proposed model achieves high accuracy rates: 89.2% for the Cleveland dataset, 89.5% for the South African dataset, 89.7% for the Z-Alizadeh Sani dataset, 90.2% for the Framingham dataset, and 91.2% for the Statlog dataset. The results obtained from the hybrid model indicate about its robustness and accuracy for medical diagnosis applicability. The proposed model outperforms various STOA of Deep Learning and Machine Learning.

The study [37] proposed hybrid optimization algorithm Hybrid Memory Improved Chameleon Swarm Algorithm (HMICSA) that aimed to ehnace the featur selection process for medical diagnosis purposes. The HMICSA consists of Chameleon Swarm Algorithm (CSA) with the Ali Baba and the Forty Thieves (AFT) algorithm that improves the overall feature selection process by the added memory features. The model was tested on 24 various medical datasets, the robustness of the proposed algorithm yield better performance than traditional approaches for feature selection for effective medical diagnosis.

The paper [38] proposed the Iterative feature selection using the Dynamic Butterfly Optimization Algorithm which is based on Interaction Maximization (IFS-DBOIM) for feature selection. The proposed hybrid algorithm was benchmark over UCI datasets repository with the mean classification accuracy of 90.43% with overall reduction in the computational complexity with consistently delivering the high accuracy rates with highly reduced number of features selected proving its robustness over the diverse data classification.

The research [39] proposed two hybrid models which were combined with the Genetic Algorithms (GA) with Stacked Autoencoder (SAE) and Convolutional Neural Network (CNN) – to classify various types of nutritional anemia. The GA was used to optimize the hyper-parameters of both the models, the hyper-tuned models thus produced the accuracy of 98.5% (GA-CNN) outperforming traditional approaches while GA-SAE achieves the accuracy of 97.02%. This demonstrates that the hybrid models approaches with Genetic algorithms is quite effective for medical image diagnosis.

The study [40] introduces a hybrid deep neural network (DNN) with an adaptive sine cosine crow search (ASCCS) for medical image data analysis of lung cancer. It uses FNLM filtering, MasiEMT-SSA segmentation, GLRLM feature extraction, and BGOA for feature selection. The proposed model was tested on LIDC-IDRI dataset, the model achieved 99.17% accuracy, 99.3% sensitivity, and 99.03% specificity, outperforming the popular machine learning algorithms. The outcomes of the proposed model demonstrate the robustness and accurately diagnosis of the lung cancer highlighting it efficiency for early stage detection of lung cancer and patient treatment planning.

The research in [41] proposes two novel hybrid Artificial Bee Colony (ABC) optimization methods to boost the classification accuracy for breast cancer diagnosis. The proposed hybrid ABC combines the clonal selection of AIS and the chemotaxis phase of BFO. These techniques majorly focuses on optimizing the parameters of the Artificial Neural Network (ANN). The propose Hybrid ABC achieves the accuracy of 99.54% on the Wisconsin Breast Cancer Dataset outperforming existing ABC and its variants and Machine Learning Algorithms. The results indicates that the hybrid techniques employed under the study boosts the accuracy and overall efficiency of the breast cancer detection and its application to medical diagnosis.

Table 2

Robust hybrid algorithms.

Algorithm name	Dataset	Result accuracy
Enhanced chaotic crow search and particle swarm optimization algorithm (ECCSPSOA) [42]	15 UCI repository datasets	89.67% average accuracy
Hybrid expert system (HES) [6]	Epilepsy diagnosis dataset	63.6% to 83.3% accuracy
Hybrid grey wolf optimizer and sine cosine algorithm (HGWOSCA) [8]	Classical and biomedical datasets	Superior performance, specific accuracy not provided
Deep belief network optimized with cuckoo search [36]	Various cardiac disease datasets	89.2% to 91.2% accuracy
Memetic algorithm-based support vector machine (M-SVM) [43]	Seven standard benchmark datasets	Up to 100% accuracy (e.g., Ovarian Cancer)
DenseNet-based deep convolutional neural network with improved coati optimization algorithm (LFR-COA-DenseNet121-BC) [45]	Comprehensive breast cancer dataset	99.97% accuracy
Adaptive hybrid-mutated differential evolution (A-HMDE) [46]	Various low and high-dimensional medical datasets	88% to 100% accuracy

The research proposed [42] a hybrid algorithm that was developed using the combination of Enhanced Chaotic Crow Search and Particle Swarm Optimization (ECCSPSOA), for feature selection in large datasets. The local optima problem of the Chaotic Crow Search was mitigated using the ECCSPSOA that improved the overall effectiveness in feature selection process hence its application in the area of medical image diagnosis. The said algorithm was analyzed over 15 UCI datasets where the proposed model achieved the overall accuracy of 89.67% outperforming various other bio-inspired algorithms.

The research [43] proposed an approach to the Support Vector Machine using the memetic algorithm (M-SVM) for parameter selection and gene selection for medical data classification. Combining the M-SVM with hybrid algorithm of Combining Emperor Penguin Optimization (EPO) with Social Engineering Optimization (SEO) outperformed SOTA whileattaining 98.82% accuracy for Leukemia, 96.44% for Colon Tumor, and 100% for Ovarian Cancer, with fewer genes and better overall feature selection.

The paper [44] proposed a hybrid model for identifying kidney stones, the hybrid model consists of CNN with the bio-inspired Flexible Dwarf Mongoose Optimization (FDMO) algorithm. To address the issues of data imbalance and variations, the model uses Look-Up correction, noise reduction, and data augmentation for preprocessing. The FDMO optimizes the CNN’s parameters and its architecture hence leading to an overall increase in the diagnosis accuracy. The model was tested on CT Kidney Dataset on which the model produced the overall accuracy of 97.12% outperforming the SOTA.

The publication [45] proposed a breast cancer diagnosis model combining a DenseNet-based CNN with an upgraded Coati Optimization Algorithm (LFR-COA). The proposed LFR-COA-Denseet121-BC model leverages the technique like arbitrary opposition-based learning, Brownian motion, and Lévy Flight to enhance CNN hyperparameters with the help of the LFR-COA leading to better convergence and robustness of the model. The proposed model outperformed the SOTA and other optimizing algorithms demonstrating its effectiveness for the breast cancer diagnosis and its applicability over other medical imaging problems.

The research [46] proposes the Adaptive Hybrid-Mutated Differential Evolution (A-HMDE) approach for overall optimized feature selection in high-dimensional medical datasets. The A-HMDE combines Spider Wasp Optimization (SWO) into Differential Evolution (DE) mutation techniques, along with adaptive control parameters, an adaptive mutation operator, and Enhanced Solution Quality (ESQ). The Differential Evolution limitations were address by the proposed A-HMDE and the proposed model was tested on UCI datasets that yield the accuracy of 88% to 100% over different medical datasets. The approach outperformed the existing feature selection algorithms as well as proves the robustness of the proposed model that can avoid the local minima and has enhanced convergence speed.

After a thorough review of the research manuscripts, the following hybrid bio-inspired algorithms have demonstrated exceptional robustness. These algorithms have been rigorously tested on various medical image datasets, consistently achieving superior accuracy compared to the state-of-the-art (SOTA) methods.

3. .Conclusion

The area of hybrid bio-inspired is a rapidly developing area of research that offers solutions to complex real-world problems with different challenges such as low amount of dataset with very high dimensionality. The hybrid approaches that were discussed in the Literature review section shows that the hybrid models are more conducive to handle complex datasets and also being robust in nature. The notable application of the hybrid bio-inspired algorithm is that their ability of feature selection particularly in the area of Medical Imaging which is the classic example of the non-linear and high dimensional data. Recent studies shows that the increasing application interest of hybrid bio-inspired algorithms with standalone bio-inspired algorithms being paired with another and with the combination of Deep Learning models, hence their is a significant study gap that have to be solved in order to fully utilize the potential of such hybrid algorithms. This literature analysis included 198 articles, after removing the 103 duplicates, 95 articles remained as per the criteria.

Finally 41 articles were selected for the study. Based on the results obtained, the future research focus on further improvement and analysis of hybrid bio-inspired algorithm based feature selection in the domain of medical imaging classification and clustering should be carried out in order to unearth the hidden potential of such algorithms. As seen in recent study the hybrid algorithms combinations have achieved better computational efficiency and mitigated the problems associated with one or the other bio-inspired algorithm or Deep Learning algorithms. Deep Learning can give powerful models that can train over complex data, thereby might help in improving the feature selection process of the bio-inspired algorithms and could lead to much optimized hybrid algorithms. These research areas can greatly increase the SOTA analysis in the domain of medical imaging, resulting in precise diagnosis and better accuracy for patients in lower income or middle income countries having limited resources at their disposals. This review serves as a foundation for steering subsequent study efforts towards these potential avenues.

4. .Future scope

As per the research undertaken in this manuscript, the Hybrid Bio-Inspired Algorithms are the way forward for the research in the feature selection in medical imaging classification and clustering. As per the latest research we have observed that hybrid solutions could uncover hidden potential and combination of such algorithms would enhance overall computational efficiency and might mitigate the shortfalls of Bio-Inspired or Deep Learning Algorithms. The combination of Bio-Inspired Algorithms and Deep Learning algorithms would lead to optimized hybrid algorithms which could significantly advance the SOTA in medical image analysis which would lead to precise diagnosis and accuracy. This review provides a foundation for directing future research efforts toward these promising avenues.

Footnotes

ORCID iDs

Anupam Kumar

Faiyaz Ahmad

Bashir Alam

References

Man

Tang

Kwong

. Genetic algorithms: concepts and applications [in engineering design]. IEEE Transactions on Industrial Electronics. 1996; 43(5): 519-534. doi: https://doi.org/10.1109/41.538609.

Beni

Wang

. Swarm Intelligence in Cellular Robotic Systems. In: Dario

Sandini

Aebischer

, editors. Robots and Biological Systems: Towards a New Bionics? Berlin, Heidelberg: Springer Berlin Heidelberg; 1993; pp. 703-712.

Kennedy

Eberhart

. Particle swarm optimization. In: Proceedings of ICNN-95 – International Conference on Neural Networks. ICNN-95. IEEE; doi: https://doi.org/10.1109/ICNN.1995.488968.

Dorigo

Di Caro

. Ant colony optimization: a new meta-heuristic. In: Proceedings of the 1999 Congress on Evolutionary Computation-CEC99 (Cat. No. 99TH8406). CEC-99. IEEE;. doi: https://doi.org/10.1109/CEC.1999.782657.

IEEE Control Systems. 2002 Jun; 22(3): 52-67. doi: https://doi.org/10.1109/MCS.2002.1004010.

Brasil

de Azevedo

Barreto

. Hybrid expert system for decision supporting in the medical area: complexity and cognitive computing. International Journal of Medical Informatics. 2001 Sep; 63(1-2): 19-30. doi: https://doi.org/10.1016/S1386-5056(01)00168-X.

Sayed

Hassanien

Azar

. Feature selection via a novel chaotic crow search algorithm. Neural Computing and Applications. 2017 Apr; 31(1): 171-188. doi: https://doi.org/10.1007/s00521-017-2988-6.

Singh

. A novel hybrid GWO-SCA approach for optimization problems. Engineering Science and Technology, an International Journal. 2017 Dec; 20(6): 1586-1601. doi: https://doi.org/10.1016/j.jestch.2017.11.001.

Woźniak

Połap

. Bio-inspired methods modeled for respiratory disease detection from medical images. Swarm and Evolutionary Computation. 2018 Aug; 41: 69-96. doi: https://doi.org/10.1016/j.swevo.2018.01.008.

10.

Jude Hemanth

Anitha

. Modified Genetic Algorithm approaches for classification of abnormal Magnetic Resonance Brain tumour images. Applied Soft Computing. 2019 Feb; 75: 21-28. doi: https://doi.org/10.1016/j.asoc.2018.10.054.

11.

Zemmal

Azizi

Sellami

Cheriguene

Ziani

AlDwairi

, et al. Particle Swarm Optimization Based Swarm Intelligence for Active Learning Improvement: Application on Medical Data Classification. Cognitive Computation. 2020 Aug; 12(5): 991-1010. doi: https://doi.org/10.1007/s12559-020-09739-z.

12.

Pasha

Latha

. Bio-inspired dimensionality reduction for Parkinson’s disease (PD) classification. Health Information Science and Systems. 2020 Mar; 8(1). doi: https://doi.org/10.1007/s13755-020-00104-w.

13.

Verma

Tiwari

. A population based hybrid FCM-PSO algorithm for clustering analysis and segmentation of brain image. Expert Systems with Applications. 2021 Apr; 167: 114121. doi: https://doi.org/10.1016/j.eswa.2020.114121.

14.

Yousri

Abd Elaziz

Abualigah

Oliva

Al-qaness

MAA

Ewees

. COVID-19 X-ray images classification based on enhanced fractional-order cuckoo search optimizer using heavy-tailed distributions. Applied Soft Computing. 2021 Mar; 101: 107052. doi: https://doi.org/10.1016/j.asoc.2020.107052.

15.

Chakraborty

Banik

Bhadra

Nandi

. Dynamically learned PSO based neighborhood influenced fuzzy c-means for pre-treatment and post-treatment organ segmentation from CT images. Computer Methods and Programs in Biomedicine. 2021 Apr; 202: 105971. doi: https://doi.org/10.1016/j.cmpb.2021.105971.

16.

Singh

Kumar

Mahmud

Kaiser

Kishore

. COVID-19 Infection Detection from Chest X-Ray Images Using Hybrid Social Group Optimization and Support Vector Classifier. Cognitive Computation. 2021 Mar; doi: https://doi.org/10.1007/s12559-021-09848-3.

17.

Patil

Hamde

. Automated detection of brain tumor disease using empirical wavelet transform based LBP variants and ant-lion optimization. Multimedia Tools and Applications. 2021 Feb; 80(12): 17955-17982. doi: https://doi.org/10.1007/s11042-020-10434-2.

18.

Dixit

Mani

Bansal

. CoV2-Detect-Net: Design of COVID-19 prediction model based on hybrid DE-PSO with SVM using chest X-ray images. Information Sciences. 2021 Sep; 571: 676-692. doi: https://doi.org/10.1016/j.ins.2021.03.062.

19.

Koyuncu

Barstuğan

. COVID-19 discrimination framework for X-ray images by considering radiomics, selective information, feature ranking, and a novel hybrid classifier. Signal Processing: Image Communication. 2021 Sep; 97: 116359. doi: https://doi.org/10.1016/j.image.2021.116359.

20.

Perumal

Narayanan

Rajasekar

SJS

. Prediction of COVID Criticality Score with Laboratory, Clinical and CT Images using Hybrid Regression Models. Computer Methods and Programs in Biomedicine. 2021 Sep; 209: 106336. doi: https://doi.org/10.1016/j.cmpb.2021.106336.

21.

Zhao

Wang

Heidari

Chen

. Performance optimization of salp swarm algorithm for multi-threshold image segmentation: Comprehensive study of breast cancer microscopy. Computers in Biology and Medicine. 2021 Dec; 139: 105015. doi: https://doi.org/10.1016/j.compbiomed.2021.105015.

22.

Doe

Smith

Roe

. Enhancing the Practical Applicability of Genetic Algorithms by Optimizing Reproduction Operators. Computers in Biology and Medicine. 2022; 146: 105383.

23.

Gopatoti

Vijayalakshmi

. CXGNet: A tri-phase chest X-ray image classification for COVID-19 diagnosis using deep CNN with enhanced grey-wolf optimizer. Biomedical Signal Processing and Control. 2022 Aug; 77: 103860. doi: https://doi.org/10.1016/j.bspc.2022.103860.

24.

Bansal

Rustagi

Kumar

. Alzheimer’s disease diagnosis based on feature extraction using optimised crow search algorithm and deep learning. International Journal of Computer Applications in Technology. 2021; 65(4): 325. doi: https://doi.org/10.1504/ijcat.2021.117272.

25.

Karunanayake

Lohitvisate

Makhanov

. Artificial life for segmentation of fusion ultrasound images of breast abnormalities. Pattern Recognition. 2022 Nov; 131: 108838. doi: https://doi.org/10.1016/j.patcog.2022.108838.

26.

Khalid

Hosny

Mirjalili

. COVIDOA: a novel evolutionary optimization algorithm based on coronavirus disease replication lifecycle. Neural Computing and Applications. 2022 Aug; 34(24): 22465-22492. doi: https://doi.org/10.1007/s00521-022-07639-x.

27.

Singh

Khanna

Thawkar

Singh

. Collaboration of features optimization techniques for the effective diagnosis of glaucoma in retinal fundus images. Advances in Engineering Software. 2022 Nov; 173: 103283. doi: https://doi.org/10.1016/j.advengsoft.2022.103283.

28.

Shaban

. Insight into breast cancer detection: new hybrid feature selection method. Neural Computing and Applications. 2022 Dec; 35(9): 6831-6853. doi: https://doi.org/10.1007/s00521-022-08062-y.

29.

Xing

Zhou

Zhao

Chen

Heidari

. Elite levy spreading differential evolution via ABC shrink-wrap for multi-threshold segmentation of breast cancer images. Biomedical Signal Processing and Control. 2023 Apr; 82: 104592. doi: https://doi.org/10.1016/j.bspc.2023.104592.

30.

Ding

Huang

Rodriguez

. Modified Locust Swarm optimizer for oral cancer diagnosis. Biomedical Signal Processing and Control. 2023 May; 83: 104645. doi: https://doi.org/10.1016/j.bspc.2023.104645.

31.

Rajeev

Pallikonda Rajasekaran

Vishnuvarthanan

Arunprasath

. A biologically-inspired hybrid deep learning approach for brain tumor classification from magnetic resonance imaging using improved gabor wavelet transform and Elmann-BiLSTM network. Biomedical Signal Processing and Control. 2022 Sep; 78: 103949. doi: https://doi.org/10.1016/j.bspc.2022.103949.

32.

Vijh

Gaurav

Pandey

. Hybrid bio-inspired algorithm and convolutional neural network for automatic lung tumor detection. Neural Computing and Applications. 2020 Sep; 35(33): 23711-23724. doi: https://doi.org/10.1007/s00521-020-05362-z.

33.

Nadimi-Shahraki

Zamani

Asghari Varzaneh

Mirjalili

. A Systematic Review of the Whale Optimization Algorithm: Theoretical Foundation, Improvements, and Hybridizations. Archives of Computational Methods in Engineering. 2023 May; 30(7): 4113-4159. doi: https://doi.org/10.1007/s11831-023-09928-7.

34.

Balasubramanian

N P

. Correlation-based feature selection using bio-inspired algorithms and optimized KELM classifier for glaucoma diagnosis. Applied Soft Computing. 2022 Oct; 128: 109432. doi: https://doi.org/10.1016/j.asoc.2022.109432.

35.

Kumar Singh

Khanna

Singh

. A novel enhanced hybrid clinical decision support system for accurate breast cancer prediction. Measurement. 2023 Nov; 221: 113525. doi: https://doi.org/10.1016/j.measurement.2023.113525.

36.

Narayan

. Cardiac disease detection using cuckoo search enabled deep belief network. Intelligent Systems with Applications. 2022 Nov; 16: 200131. doi: https://doi.org/10.1016/j.iswa.2022.200131.

37.

Braik

Hammouri

Awadallah

Al-Betar

Khtatneh

. An improved hybrid chameleon swarm algorithm for feature selection in medical diagnosis. Biomedical Signal Processing and Control. 2023 Aug; 85: 105073. doi: https://doi.org/10.1016/j.bspc.2023.105073.

38.

Tiwari

Chaturvedi

. A hybrid feature selection approach based on information theory and dynamic butterfly optimization algorithm for data classification. Expert Systems with Applications. 2022 Jun; 196: 116621. doi: https://doi.org/10.1016/j.eswa.2022.116621.

39.

Kilicarslan

, Celik

. Sahin Hybrid models based on genetic algorithm and deep learning algorithms for nutritional Anemia disease classification. Biomedical Signal Processing and Control. 2021 Jan; 63: 102231. doi: https://doi.org/10.1016/j.bspc.2020.102231.

40.

Surendar

Bala Ponni

. Diagnosis of lung cancer using hybrid deep neural network with adaptive sine cosine crow search algorithm. Journal of Computational Science. 2021 Jul; 53: 101374. doi: https://doi.org/10.1016/j.jocs.2021.101374.

41.

Punitha

Stephan

Gandomi

. A Novel Breast Cancer Diagnosis Scheme With Intelligent Feature and Parameter Selections. Computer Methods and Programs in Biomedicine. 2022 Feb; 214: 106432. doi: https://doi.org/10.1016/j.cmpb.2021.106432.

42.

Adamu

Abdullahi

Junaidu

Hassan

. An hybrid particle swarm optimization with crow search algorithm for feature selection. Machine Learning with Applications. 2021 Dec; 6: 100108. doi: https://doi.org/10.1016/j.mlwa.2021.100108.

43.

Baliarsingh

Ding

Vipsita

Bakshi

. A memetic algorithm using emperor penguin and social engineering optimization for medical data classification. Applied Soft Computing. 2019 Dec; 85: 105773. doi: https://doi.org/10.1016/j.asoc.2019.105773.

44.

Liu

Ghadimi

. Hybrid convolutional neural network and Flexible Dwarf Mongoose Optimization Algorithm for strong kidney stone diagnosis. Biomedical Signal Processing and Control. 2024 May; 91: 106024. doi: https://doi.org/10.1016/j.bspc.2024.106024.

45.

Emam

Houssein

Samee

Alohali

Hosney

. Breast cancer diagnosis using optimized deep convolutional neural network based on transfer learning technique and improved Coati optimization algorithm. Expert Systems with Applications. 2024 Dec; 255: 124581. doi: https://doi.org/10.1016/j.eswa.2024.124581.

46.

Mostafa

Khedr

Al Aghbari

Afyouni

Kamel

Ahmed

. An adaptive hybrid mutated differential evolution feature selection method for low and high-dimensional medical datasets. Knowledge-Based Systems. 2024 Jan; 283: 111218. doi: https://doi.org/10.1016/j.knosys.2023.111218.

Hybrid bio-inspired computing in medical image data analysis: A review

Abstract

OBJECTIVES

METHOD

RESULT

CONCLUSION

Keywords

1. .Introduction

1.1. .Scope of discussion

1.2. .Our contribution

Table 1 BALO-S and BALO-V binary ant lion optimization. Algorithm name Dataset Result accuracy Hybrid genetic algorithm Nutritional anemia datasets 91.2% BALO-S Brain MR images dataset 97.22% BALO-V Brain MR images dataset 97.22%

4. .Future scope

Footnotes

ORCID iDs

References

Table 1
BALO-S and BALO-V binary ant lion optimization.

Algorithm name Dataset Result accuracy

Hybrid genetic algorithm Nutritional anemia datasets 91.2%

BALO-S Brain MR images dataset 97.22%

BALO-V Brain MR images dataset 97.22%