Swarm intelligence based clustering technique for automated lesion detection and diagnosis of psoriasis

https://doi.org/10.1016/j.compbiolchem.2020.107247Get rights and content

Highlights

Abstract

Background

In psoriasis skin disease, psoriatic cells develop rapidly than the normal healthy cells. This speedy growth causes accumulation of dead skin cells on the skin’s surface, resulting in thick patches of red, dry, and itchy skin. This patches or psoriatic skin legions may exhibit similar characteristics as healthy skin, which makes lesion detection more challenging. However, for accurate disease diagnosis and severity detection, lesion segmentation has prime importance. In that context, our group had previously performed psoriasis lesion segmentation using the conventional clustering algorithm. However, it suffers from the constraint of falling into the local sub-optimal centroids of the clusters.

Objective

The main objective of this paper is to implement an optimal lesion segmentation technique with aims at global convergence by reducing the probability of trapping into the local optima. This has been achieved by integrating swarm intelligence based algorithms with conventional K-means and Fuzzy C-means (FCMs) clustering algorithms.

Methodology

There are a total of eight different suitable combinations of conventional clustering (i.e., K-means and Fuzzy C-means (FCMs)) and four swarm intelligence (SI) techniques (i.e., seeker optimization (SO), artificial bee colony (ABC), ant colony optimization (ACO) and particle swarm optimization (PSO)) have been implemented in this study. The experiments are performed on the dataset of 780 psoriasis images from 74 patients collected at Psoriasis Clinic and Research Centre, Psoriatreat, Pune, Maharashtra, India. In this study, we are employing swarm intelligence optimization techniques in combination with the conventional clustering algorithms to increase the probability of convergence to the optimal global solution and hence improved clustering and detection.

Results

The performance has been quantified in terms of four indices, namely accuracy (A), sensitivity (SN), specificity (SP), and Jaccard index (JI). Among the eight different combinations of clustering and optimization techniques considered in this study, FCM + SO outperformed with mean JI = 0.83, mean A = 90.89, mean SN = 92.84, and mean SP = 88.27. FCM + SO found statistical significant than other approaches with 96.67 % of the reliability index.

Conclusion

The results obtained reflect the superiority of the proposed techniques over conventional clustering techniques. Hence our research development will lead to an objective analysis for automatic, accurate, and quick diagnosis of psoriasis.

Introduction

Psoriasis is a type of skin disease which affects human being irrespective of gender and age. At present, approximately 1–3 % of the worldwide population is affected by psoriasis (Huerta et al., 2007; Sander and Norris, 1993; Adam, 1980; Yee et al., 1999). In psoriatic patients, skin cells develop rapidly with an average of 4–5 days as compared to healthy subjects (i.e., 25–30 days) (Shrivastava et al., 2015a). It can affect the entire part of the body, but the most severely affected parts are scalp, joints, and nails (Roenigk and Maibach, 1985; Camisa, 2004). Moreover, disease brings embarrassment in public and reduced value of life to the patients (Henseler, 1998; Taur et al., 2006).

This disease requires routine and disciplined treatment with proper attention. The existing techniques of treatment implemented by dermatologists are subjective in the process, which involves visual inspection and customary testing (Taur et al., 2006; Feldman and Krueger, 2005). However, it is difficult to predict the severity of psoriasis disease by the subjective analysis, which may vary based on the expertise and skills of a medical expert. Moreover, the color diversity of the lesion region makes the detection task quite challenging (Xie and Bovik, 2013). Hence, the subject analysis is unreliable and time-consuming due to probable inter and intra-observer inconsistency (Feldman and Krueger, 2005). Thus an objective analysis is essential for efficient, correct and fast diagnosis

In view of that, an automatic and effective computer-aided diagnosis (CADx) system for severity detection of psoriasis lesions was developed by Shrivastava et al. (2015a, 2015b,c; Shrivastava et al., 2016b,c; Shrivastava et al., 2017. They describe the features which contribute more towards identifying the severity and formation of initial clusters. Nidhal et al. Nidhal et al. (2010) and Taur (Taur, 2003) proposed a classifier based technique to segment the psoriasis lesion. Bogo et al. (2012) applied a contour-based approach for the psoriasis segmentation. Other clustering-based (Li-Hong and Ming-Ni, 2011; Shrivastava and Londhe, 2015) and model-based (Taur et al., 2006; Guoli et al., 2013) methods have been proposed for the segmentation of psoriasis images. However, the results of most of the presented works (Shrivastava et al. (2015a); Taur et al., 2006; Shrivastava et al. (2015b); Shrivastava et al., 2015c, a; Shrivastava et al., 2016b, c; Shrivastava et al., 2017; Taur (2003); Bogo et al., 2012; Li-Hong and Ming-Ni, 2011; Shrivastava and Londhe, 2015) are not reliable as experimentations have been carried out with small data size (Taur, 2003; Bogo et al., 2012; Li-Hong and Ming-Ni, 2011; Shrivastava and Londhe, 2015). Validation on a larger dataset with varying degree of severity allows the applicability of the technique for the design of a reliable and generalized CADx system. Most of the earlier proposed CADx systems involves a number of feature extraction and selection processes which consume more time and computationally complex. Hence, they required to be fed with accurately segmented lesion images for improved performance. In line with that, we have proposed an effective lesion segmentation method based on optimization for better detection accuracy for which only two sets of features namely a (green-red) and b (blue-yellow) have been used from CIE Lab color space.

Few researchers have also applied optimization techniques based on evolutionary computation to perform clustering and image segmentation. A dynamic clustering approach based on particle swarm optimization (PSO) was proposed by Omran et al. Omran et al. (2006) and Ma et al. Ma et al. (2011). In (Sag and Unkas, 2012) Sag et al. have applied artificial bee colony (ABC) algorithm. Akay Akay (2013) exploited the search abilities of PSO and ABC algorithms on multi-level thresholding. Image segmentation using evolutionary algorithms was proposed by Bhandarkar et al. (Bhandarkar and Zhang, 1999) and Veenman et al. Veenman et al. (2003). All the above-reported algorithms Omran et al. (2006); Veenman et al. (2003) are implemented on standard color images. Ghosh et al. (Ghosh and Mitchell, 2006) and McIntosh et al. (McIntosh and Hamarneh, 2006) used a genetic algorithm (GA) for medical image segmentation. However, none of the above-reported works have implemented the algorithm on psoriasis image segmentation.

Most of the clustering techniques (Li-Hong and Ming-Ni, 2011; Shrivastava and Londhe, 2015; Su and Chou, 2001; Chuang et al., 2006) used for lesion detection forms the clusters by repositioning the data points to the most neighboring centroid and updating the cluster centroids. The accuracy of lesion detection using conventional clustering techniques is quite low due to (1) Convergence to local optima centroids instead of global optima; (2) Random initialization of clusters; (3) Initial assumption of the number of clusters (Kalyani and Swarup, 2011; Ahmed et al., 2002). As compared to classical gradient-based techniques used in conventional clustering approaches, evolutionary algorithm based clustering increases the probability of global convergence with reduced dependence on initial parameterization.

The present work aims at improving the accuracy of lesion detection by overcoming the limitations of conventional clustering techniques (Li-Hong and Ming-Ni, 2011; Shrivastava and Londhe, 2015; Su and Chou, 2001; Chuang et al., 2006; Kalyani and Swarup, 2011; Ahmed et al., 2002). In this regard, we propose an approach based on the hybridization of swarm intelligence (SI) based optimization algorithms, i.e., seeker optimization (SO), ABC, ant colony optimization (ACO) and PSO with two conventional clustering techniques (K-means and Fuzzy C-means). It provides a significant improvement in accuracy of psoriasis lesion detection by identifying image clusters, having better optimal centroids and which may be global optimum centroids. The proposed approach compares the ability of different SI based optimization algorithms (SO, ABC, ACO, and PSO) individually. The centroids obtain from conventional clustering techniques is considered as the initial solution for the SI algorithms to fine-tune the centroids to improve the probability of convergence to the global optimal centroids. The proposed approach has been applied to a set of 780 images, and its effectiveness has been evaluated in terms of different matrices, i.e., Jaccard index, sensitivity, specificity, and accuracy. In summary, the major contributions of this paper are: (i) implementation of swarm intelligence algorithms to overcome the local search problem in conventional clustering algorithms; (ii) effective psoriasis lesion detection using four swarm intelligence techniques (PSO, Ant Colony, ABC and Seeker Optimization); (iii) effective analysis of segmentation method using four quantitative metrics (Accuracy, Jaccard Index, Specificity and sensitivity). (iv) proposed methods are implemented on a more extensive data set of 780 psoriasis images.

Section snippets

Methodology

In this paper, the segmentation of psoriasis images has been outlined as an optimization problem. In this regard, an objective function has been framed which seeks to minimize the Euclidian distance between pixels and centroid of each cluster as a function of centroids (Cen) for a given number of clusters K. For the present task of psoriasis lesion detection, the number of clusters is fixed at two. With two clusters, we are able to segregate healthy and psoriasis lesions. Here the objective is

Results

The above-proposed methodology of lesion segmentation using swarm intelligence based clustering has been implemented on large self-generated psoriasis image dataset in the following section. The experimentations are performed using the criss-cross combinations of selected swarm intelligence techniques PSO, ACO, ABC and SO with conventional clustering methods K-means and FCM, as per the above-proposed methodology.

Discussion

From the results, it can be observed that the conventional clustering algorithms FCM and K-means are able to attain a mean accuracy of 82.63 (with SD = 5.98) and 81.29 (with SD = 6.03) respectively. A higher accuracy relates to the better detection of psoriasis lesion with reference to the ground truth. After hybridization with swarm intelligence techniques, the lesion detection in terms of accuracy is enhanced by 10.35 % (90.89 with SD = 3.01 in FCM + SO). Again among the eight hybridization

Conclusion

This paper presents a novel psoriasis image segmentation technique, which is an automatic clustering-based segmentation procedure for lesion detection. Here, we have integrated the conventional clustering algorithms (K-means and FCM) with the well-known SO, ABC, ACO, and PSO swarm intelligence based algorithms. Our proposed technique has been implemented and tested on a dataset of 780 psoriasis images from 74 different patients. The results show that the new segmentation techniques perform much

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Declaration of Competing Interest

None declared.

Acknowledgments

The authors would like to thank the psoriasis patients who participated in this study.

References (44)

  • V.K. Shrivastava et al.

    A novel and robust Bayesian approach for segmentation of psoriasis lesions and its risk stratification: inter-comparison of nine systems

    Comp. Methods Programs Biomed.

    (2017)
  • F. Xie et al.

    Automatic segmentation of dermoscopy images using self-generating neural networks seeded by genetic algorithm

    Patt. Recogn.

    (2013)
  • B.A. Adam

    Psoriasis in hospital population

    Med. J. Malaysia

    (1980)
  • M.N. Ahmed et al.

    A modified fuzzy c-means algorithm for bias field estimation and segmentation of MRI data

    IEEE Trans. Med. Imaging

    (2002)
  • S. Bansal et al.

    Color image segmentation using CIELab color space using ant Colony optimization

    Int. J. Computer Appl.

    (2011)
  • S.M. Bhandarkar et al.

    Image segmentation using evolutionary computation

    IEEE Trans. Evol. Comput.

    (1999)
  • F. Bogo et al.

    Psoriasis segmentation through chromatic regions and geometric active contours

    International Conference of the IEEE EMBS

    (2012)
  • C. Camisa

    Handbook of Psoriasis

    (2004)
  • K.S. Chuang et al.

    Fuzzy C-Means Clustering with Spatial Information for Image Segmentation, in Computerized Medical Imaging and Graphics, 30, 9, 15

    (2006)
  • C. Connolly et al.

    A study of efficiency and accuracy in the transformation from RGB to CIELAB color space

    IEEE Trans. Image Process.

    (1997)
  • M. Dorigo et al.

    Positive Feedback as a Search Strategy, Tech. Report 91-016, Department Electronic

    (1991)
  • S.R. Feldman et al.

    Psoriasis assessment tools in clinical trials

    Ann. Rheumatic Dis.

    (2005)
  • Cited by (23)

    • Automated psoriasis lesion segmentation from unconstrained environment using residual U-Net with transfer learning

      2021, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      All these research works were done with a self-prepared psoriasis dataset as the standard psoriasis data is not available publically. Most of the research studies [19–30] were done with the cropped and small patches of skin images having single lesions. The neuro-fuzzy classifier was proposed in [19] for psoriasis lesion segmentation by locating homogeneous regions.

    • Research on classification diagnosis model of psoriasis based on deep residual network

      2021, Digital Chinese Medicine
      Citation Excerpt :

      However, only a small number have been diagnosed, mainly owing to a lack of diagnostic ability of grassroots doctors or hospitals. At the “Popular Science Activities of World Psoriasis Day” held in 2018, experts considered that artificial intelligence (AI) technology combined with big data can provide patients with more intuitive auxiliary diagnosis of psoriasis [5-8]. Psoriasis cannot be completely cured.

    View all citing articles on Scopus
    View full text