Medical media analytics via ranking and big learning: A multi-modality image-based disease severity prediction study

doi:10.1016/j.neucom.2015.07.148

Neurocomputing

Volume 204, 5 September 2016, Pages 125-134

https://doi.org/10.1016/j.neucom.2015.07.148 Get rights and content

Abstract

Medical media analytics receives vast popularity nowadays because of its effectiveness in improving the performance of diverse health-care applications. In this study, the essential disease severity prediction problem in medical media analytics is investigated and a computer-aided diagnosis (CAD) strategy based on ranking and learning techniques is presented to tackle the disease severity prediction task. To be specific, two types of magnetic resonance images (MRI), including T1-weighted images as anatomic MRI and arterial spin labeling (ASL) images as functional MRI, are incorporated as multi-modality images to provide image-based information for dementia disease severity prediction in this study. There are two main steps composed of the whole CAD strategy. First, the problem of partial volume effects (PVE) mainly caused by signal cross-contamination due to pixel heterogeneity and limited spatial resolution of ASL is focused. Conventional regression-based PVE correction methods are discussed and their inherent problems of blurring and brain details loss in correction results, which prevents the actual brain atrophy being revealed, are studied. A pixel-based PVE correction method, which only counts on single pixel information and formulates the PVE correction problem as a constrained optimization problem solved via the split-Bregman algorithm, is presented to solve the problem. Second, ranking and learning techniques are incorporated based on multi-modality images after performing PVE correction for dementia disease severity prediction. Technically, a conventional discrete position-based ranking evaluation measure is approximated and its surrogated continuous form is optimized via gradient ascend for ranking functions learning. A large database composed of multi-modality images acquired from 320 real patients is utilized for experimental evaluation. Extensive experiments and comprehensive statistical analysis are carried out to demonstrate the superiority of the introduced CAD strategy with comparison to several existing ones. Promising results are reported from the statistical perspective.

Introduction

It is widely acknowledged that social media analytics receives vast popularity nowadays because of its latent capability of measuring, analyzing, interpreting diverse interactions and associations between a large group of populations, as well as its powerful capacity in uncovering and predicting individuals sentiment [1]. Medical social media analytics, on the other hand, specializes in health-care utilizations and is particularly beneficial for patients׳ disease prediction and diagnosis, with the help of diverse types of information (e.g., either text-based or image-based) as well as big learning techniques (e.g., deep learning and ranking). In this study, the essential disease prediction problem in medical social medial analytics is focused. Novel ranking and learning techniques are introduced and incorporated for Alzheimer׳s Disease (AD) prediction based on multi-modality image-based information.

AD, the most common form of dementia disease, is often diagnosed in patients over 60 years old, and generally regarded as one of the five most serious non-communicable diseases (i.e. others include cardiovascular disease, cancer, diabetes and chronic lung disease) in the whole world reported by the World Health Organization [2]. According to a population-based study conducted by the United Nations, there are already over 26.6 million AD patients diagnosed globally [3], and 1 in 85 people all around the world is predicted to be suffering from AD by the year 2050 [4]. In China, the number of AD patients is believed to exceed 10 million. Accurate diagnosis and timely treatment is essential to delay the onset and progression of AD [4], and it can be realized in medical social media analytics with the help of ranking and learning techniques.

In order to diagnose the progression of dementia disease accurately, various medical imaging modalities have been investigated and utilized. Among them, Magnetic Resonance Imaging (MRI) is a powerful imaging tool and receives vast popularity because it is free of radiation exposure, compared with other conventional imaging tools such as Computed Tomography (CT) and Positron Emission Tomography (PET), for patients safety issues. For MRI, a variety of imaging techniques are proposed in the last decades, and many of them can be categorized into two types, anatomical MRI (aMRI) and functional MRI (fMRI) [5]. It is commonly seen for aMRI to be utilized in both clinical diagnosis and academic research for obtaining the anatomy of scanned patients, while fMRI receives more and more research interests nowadays by providing extra information about scanned patients, which may become obscure in aMRI, through detecting associated changes in blood flow of scanned patients [5]. Arterial Spin Labeling (ASL), which is an emerging fMRI technique, receives increasing attention in dementia diagnosis studies recently [6]. Compared with other conventional fMRI techniques, such as Blood Oxygen Level Dependent (BOLD), ASL requires no injection of external contrast enhancement agent (e.g., gadolinium) into patients while being scanned. Thus, unfavored anaphylactoid reactions on patients [7] can be totally avoided in ASL images, making it absolutely safe and more favored for dementia diagnosis at present.

Technically, an ASL image is produced by two types of images: a label image and a control image. Their acquisition steps are illustrated in Fig. 1. The yellow region 2 in Fig. 1a and the green region 4 in Fig. 1b describe the same Region-of-Interest (ROI), in which ASL images are acquired. The purple region 1 in Fig. 1a represents an area where arterial blood water is magnetically labeled via a 180° Radio-Frequency (RF) inversion pulse. In this way, water molecules within the arterial blood are magnetically labeled and utilized as the “tracer”, instead of the conventional injected contrast enhancement agent mentioned above. Label images are taken when labeled blood water flows into the ROI, and example label images from the transverse view acquired from one patient in this study are displayed in Fig. 1a. For control images, the blood water is not magnetically labeled, and control images are taken at the same ROI directly. Example control images of the same patient are displayed in Fig. 1b. Although label and control images look similar towards each other, certain differences exist between them and an ASL image is produced as their difference (i.e., using a control image minus a corresponding label image) therein, and example ASL of the same patient is illustrated in Fig. 1c. Generally speaking, the Cerebral Blood Flow (CBF) on each pixel of ASL is proportional towards its ASL signal, and brain atrophy within particular brain regions of demented patients can be revealed by low measured CBF within those regions, compared with the ones of ordinary people, reflected in ASL.

Although ASL is a promising bio-marker for disease diagnosis and progression analysis in dementia, the problem of Partial Volume Effect (PVE) should be carefully tackled. PVE is generally defined as the loss of apparent activity in small objects because of the limited resolution of an imaging system [8]. In ASL, since its spatial resolution is not high (i.e., it can be perceived by example images in Fig. 1), pixels in ASL images containing various tissues of Gray Matter (GM), White Matter (WM) and Cerebro-Spinal Fluid (CSF) are likely to be assigned with under-estimated ASL signal and low CBF quantities, which reflects the loss of apparent activity in ASL because of the problem of PVE. In order to correct PVE, there are already several studies proposed in recent years [9], [10], [11], and the regression-based method receives much popularity among them [9]. Its main idea is to formulate the PVE correction problem into indefinite equations, and solve them with the aid of neighboring pixels, when dealing with PVE on one single pixel. However, its shortcoming is also obvious. Neighboring pixels are usually indispensable for PVE correction on each single pixel of ASL, making blurring and loss of brain details inevitable in correction results of this method [9]. A case in point is illustrated in the 1st row of Fig. 2. CBF calculated based on those corrected ASL is inaccurately low, thus brain atrophy revealed by CBF in particular brain regions from those corrected results cannot be accurate enough to reveal patients with dementia disease, resulting in diagnosis errors thereafter. Therefore, in order to enable ASL a reliable indicator for the following dementia disease diagnosis, the problem of PVE needs to be properly tackled.

After PVE correction on ASL is conducted, the next critical step in dementia diagnosis is to predict the dementia disease severity based on corrected ASL of each patient. Dementia studies incorporating ASL only begin to emerge in recent years [6], [12], [13], and most of them mainly concentrate on verifying ASL as a novel indicator in identifying dementia disease, with comparison to other previously well-established imaging modalities. For the majority of contemporary dementia disease diagnosis studies, they mainly rely on conventional pattern recognition tools [14], [15], [16]. For instance, cortical thickness maps are generated from aMRI and Support Vector Machine (SVM) is employed to differentiate Mild Cognitive Impairment (MCI) from AD in [14]. In [15], the curse-of-dimension problem commonly found in pattern recognition studies is investigated for dementia diagnosis, and ensemble classifiers are constructed via sparse encodings for dementia disease prediction. In [16], local volumetric measurements obtained from aMRI are fed into hierarchical networks to discern MCI patients from AD patients. It can be summarized from existing studies that dementia disease prediction is often considered as either a classification or a regression problem.

The dementia disease prediction task in this study is, however, regarded as a ranking problem. Ranking is an emerging approach in machine learning and information retrieval in recent years [17], [18]. Generally speaking, ranking is often associated with learning techniques, and both of them are employed as two critical steps in a sequence as follows [19], [20], [21]. For learning, its characteristics in the ranking process are described as follows. Provide a set of m image lists $d^{(j)} = {d_{1}^{(j)}, d_{2}^{(j)}, \dots, d_{m^{(j)}}^{(j)}}$ with their corresponding relevance $r^{(j)} = {r_{1}^{(j)}, r_{2}^{(j)}, \dots, r_{m^{(j)}}^{(j)}}$ , $j = 1, \dots, m$ ; $m^{(j)}$ denotes the number of images within the list $d^{(j)}$ , a ranking function f is learned from these training data. Generally speaking, ranking function f is defined in terms of each individual image: $f (d_{i}^{(j)})$ , $i = 1, \dots, m^{(j)}$ with its output as the score of each image. The learned ranking function will be used to sort the image collection in the ranking step. For ranking, provided a list of n images $d = {d_{1}, d_{2}, \dots, d_{n}}$ , the purpose of ranking is to sort images within the list in a/an descending/ascending order of relevance measured by the score of each image calculated from the learned ranking function f.

In this paper, we, computer scientists and clinicians working closely together, introduce a computer-aided diagnosis (CAD) strategy for dementia diagnosis based on multi-modality MRI images (including both aMRI and ASL) via ranking and learning techniques in medical social media analytics. The usage of multi-modality MRI images in this study is explicitly explained as follows. aMRI is popular in conventional MRI studies because of its high spatial resolution, so that fine structures within human brains can be easily observed within it. However, aMRI cannot reveal functional activities of human brains, which are capable to be measured via the introduced fMRI tool, i.e., ASL, instead. Thus, multi-modality MRI images composed of both aMRI and ASL are beneficial in providing valuable visual information regarding both fine structures and functional activities of human brains. For this CAD strategy, it is composed of two steps. First, a pixel-based PVE correction method only utilizing single-pixel information for its own PVE problem is introduced. Problems of blurring and brain details loss commonly seen in correction results of conventional PVE correction methods can be properly tackled within this method. Second, a dementia disease severity prediction method based on ranking and learning techniques is elaborated to fulfill the disease diagnosis task based on corrected ASL images and its corresponding aMRI. The organization of this paper is depicted as follows. The pixel-based PVE correction method is first elaborated in Section 2. After that, the multi-modality image-based disease prediction method based on ranking and learning is described in Section 3. Comprehensive statistical experiments are conducted to evaluate the performance of the new strategy for dementia disease, with comparison to several conventional diagnosis tools in Section 4. Finally, the conclusion of this study is drawn and future directions are suggested in Section 5.

Section snippets

Pixel-based partial volume correction: a prerequisite step in ASL images processing

The PVE correction problem in ASL images can be mathematically described as follows. Given a single pixel i in an ASL image, its ASL signal $\frac{Δ M}{M_{C}}$ can be represented as follows: $\frac{Δ M}{M_{C}} = \frac{P_{GM} \cdot Δ M_{GM} + P_{WM} \cdot Δ M_{WM}}{P_{GM} \cdot M_{GM}^{C} + P_{WM} \cdot M_{WM}^{C} + P_{CSF} \cdot M_{CSF}^{C}}$ where P_GM, P_WM, P_CSF denote the fractional GM, WM and CSF tissue volume on pixel i respectively, which can be obtained from a pre-requisite brain segmentation step via the SPM toolbox [22]; $M_{⋆}^{C}$ and $Δ M_{⋆}$ indicate the control magnetization and the ASL magnetization caused by

Dementia disease severity prediction using ranking and learning

After PVE correction on ASL images is carried out, the next critical step is to predict the dementia disease severity of patients based on their corrected ASL images as well as corresponding aMRI in this study. A method based on ranking and learning techniques is presented in this section. Generally speaking, the main purpose of ranking is to sort a list of images according to their disease severities described by a ranking function, and this ranking function can be determined via a learning

Data description

In order to demonstrate the superiority of the introduced CAD strategy for dementia disease prediction, clinical data obtained from 320 patients in the affiliated hospital of Nanchang University is utilized to construct a database for experimental evaluation in this study. Patients with different dementia disease progressions, including AD, MCI and NCI (i.e., Non-Cognitive Impairment), are included. To be specific, there are 107 AD patients, 107 MCI patients and 106 NCI patients. All images of

Conclusion

Medical social media analytics is beneficial in health-care nowadays for patients disease diagnosis with the aid of multi-modality images. In this study, a novel computer-aided diagnosis strategy based on aMRI and ASL images is introduced for dementia disease severity prediction. Two main steps are made up of the whole CAD strategy, including a pixel-based PVE correction method and a disease severity prediction method based on ranking and learning techniques. Comprehensive statistical

Acknowledgment

The authors would like to acknowledge Grants 61403182 and 61363046 approved by the National Natural Science Foundation of China, the Grant [2014]1685 approved by the Scientific Research Foundation for Returned Overseas Chinese Scholars, Ministry of Education, China, as well as the 2015 Provincial Young Scientist Program 20153BCB23029 approved by the Jiangxi Provincial Department of Science and Technology, China.

References (31)

R. Brookmeyer et al.
Forecasting the global burden of Alzheimer׳s disease
Alzheimer׳s Dement.
(2007)
Z. Wang et al.
Arterial spin labeled mri in prodromal Alzheimer׳s diseasea multi-site study
NeuroImage: Clinical
(2013)
M. Liu et al.
Ensemble sparse classification of Alzheimers disease
NeuroImage
(2012)
M. Sponder, Social Media Analytics: Effective Tools for Building, Interpreting, and Using Metrics, 1st edition,...
World Health Organization (WHO), The Top 10 Causes of Death,...
United Nations (UN), World Population Prospects,...
G. Burghart, C. Finn, Handbook of MRI Scanning, The Mosby Press, Missouri, United States,...
K. Malpass
Alzheimer diseasearterial spin-labeled mri for diagnosis and monitoring of AD
Nat. Rev. Neurol.
(2012)
K. Murphy et al.
Adverse reactions to gadolinium contrast mediaa review of 36 cases
Am. J. Roentgenol.
(1996)
Partial Volume Imaging, 〈http://en.wikipedia.org/wiki/Partial_volume_(imaging)〉,...

I. Asllani et al.

Regression algorithm correcting for partial volume effects in arterial spin labeling MRI

Magn. Reson. Med.

(2008)

Y. Du et al.

Partial volume effect compensation for quantitative brain SPECT imaging

IEEE Trans. Med. Imaging

(2005)

K. Erlandsson et al.

A review of partial volume correction techniques for emission tomography and their applications in neurology, cardiology and oncology

Phys. Med. Biol.

(2012)

Y. Chen et al.

Voxel-level comparison of arterial spin-labeling perfusion MRI and FDG-PET in Alzheimer disease

Neurology

(2011)

C. Wee et al.

Prediction of Alzheimers disease and mild cognitive impairment using cortical morphological patterns

Hum. Brain Map.

(2013)

Cited by (29)

A novel locally-constrained GAN-based ensemble to synthesize arterial spin labeling images
2022, Information Sciences
Arterial spin labeling (ASL) images that are capable to quantitatively measure the cerebral blood flow receive increasing research attention in recent dementia diagnosis studies. However, this important imaging modality is unfortunately not commonly seen in many well-established image-based dementia datasets. Hence, synthesizing ASL images to supplement the important modality in these datasets for further improving the accuracy of dementia diseases diagnosis is quite important and valuable. In this study, a novel locally-constrained generative adversarial networks (GAN)-based ensemble is introduced to fulfill the ASL image synthesis task for improving the dementia diseases diagnosis performance. Technically, new attention-based feature pyramid-GAN models are designed as local models of the novel ensemble. Also, multi-Gaussian-distributed noise is generated from a new flow-based generative model and utilized in medical image synthesis, for the first time. Experiments have been conducted to reveal the effectiveness of the novel GAN ensemble. Comparisons between the novel GAN ensemble and many other state-of-the-art methods in medical image synthesis have been carried out. Statistical analyses have suggested that, accuracies of dementia diseases diagnosis can be significantly improved with the help of the novel GAN ensemble, which brings about $41.62 %$ performance improvement based on a 355-demented-patient dataset and approximately $25 %$ performance improvement from the well-known ADNI-1 dataset.
Multi-criterion decision making-based multi-channel hierarchical fusion of digital breast tomosynthesis and digital mammography for breast mass discrimination
2021, Knowledge-Based Systems
Citation Excerpt :
Previous clinical studies [12–15] have shown that combining DBT and FFDM enhances cancer detection and classification rate in mammography screening. In recent years, computer-aided diagnosis (CAD) has been actively investigated as an alternative and complementary approach to conventional reading by radiologists [16], which is susceptible to inter-observer bias and demands high level of expertise. In the field of breast disease diagnosis, hand-crafted features (e.g., histogram features and texture features [17,18] have been widely applied for CAD-based analysis such as breast density classification [19] and breast cancer classification [20–24].
Multifaceted features decoded from mammographic images may describe various perspectives of the breast mass heterogeneity, in this study, we aimed to explore a methodology to effectively integrate multifaceted mass representations extracted from the digital breast tomosynthesis (DBT) and full-field digital mammography (FFDM) to enhance breast cancer discrimination. A novel multi-criterion decision making-based multi-channel fusion (MDMF) framework was proposed to fuse different breast mass representations processed in multi-channels built on the deep convolutional neural network (DCNN) and the multilayer perceptron (MLP) at the decision level. A hierarchical framework (HFMM) was also developed for multi-modality images and multi-channel fusion to integrate multimodality information from DBT and FFDM. We retrospectively collected 441 patients with both DBT and FFDM, and the regions of interest (ROIs) covering the malignant, benign, and normal tissues were extracted for validation. The MDMF achieved the area under the receiver operating characteristic curve (AUC) of 93.14%, 91.30%, 97.35% (FFDM) and 93.79%, 95.16%, 99.31% (DBT) respectively for the malignant, benign and normal mass. While the HFMM further boosted the performance to AUC of malignant 94.14%, benign 95.42% and normal mass 99.56% The matthews correlation coefficient (MCC) were 73.15% and 81.02% for FFDM and DBT accomplished by MDMF, and enhanced to 81.72% when integrating the multimodality information from DBT and FFDM via the proposed HFMM. The experimental results suggested that the proposed HFMM achieved superior discriminative performance when compared with the benchmark classification algorithms and fusion architectures, rendering it a practical tool for breast mass discrimination in breast cancer screening.
Classification and Identification of Primitive Kharif Crops using Supervised Deep Convolutional Networks
2020, Sustainable Computing: Informatics and Systems
Citation Excerpt :
Barbedo [4] proposed image translation and restoration techniques for detection, classification, and quantification of plant diseases. Huang et al. [5] proposed multi-modality image-based disease severity prediction study using big learning and analytics prediction. Phadikar et al. [6] preferred pattern recognition techniques for automated rice disease identification and achieved an accuracy of 70%.
The severity of diseases and threats found in different crop varieties is one of the primary causes of degradation in the agricultural economy. Early detection and disease diagnosis in crops will facilitate farmers to improve their livelihood and mankind.
This study aimed to develop method for disease identification in several seasonal crops during their early stages using deep learning architectures i.e. convolutional neural networks (CNN) and compare the feasibility, accuracy and performance of the proposed network with conventional feature extraction techniques like support vector machine, k-nearest neighbor, genetic algorithm, and artificial neural networks.
This study preferred a database of 600 images i.e. 200 images of individual crop varieties which are labeled with 10 kinds of crop diseases. Each crop varieties have two different kinds of classes i.e. health crop and rusty crop. The CNNs are trained in such a manner that it will be able to detect diseases from infected crop varieties.
Different convolution filters and pooling types of different sizes are used in the proposed work. Max pooling with a filter size of 32*32*3 achieves the accuracy of 92%. Average pool size with a convolution filter size of 64*64*3 achieved maximum accuracy of 93.7% and gains the better results in comparison to other machine learning and feature extraction models.
The contribution of the proposed work could be summarized as: (i) obtained results shows improvement in the feasibility and performance of CNN over other machine learning models. (ii) High performance shows the immediate crop disease identification ability of deep learning techniques over the different feature extraction models.
Aspect-based opinion ranking framework for product reviews using a Spearman's rank correlation coefficient method
2018, Information Sciences
Citation Excerpt :
This ranking is not based on language models (LM). Huang et al. [39] investigated dementia's disease severity prediction using a ranking and presented learning techniques for the disease severity prediction task. Marrese-Taylor et al. [9] presented an extension of Bing Liu's aspect-based opinion mining approach in tourism product reviews for sentiment classification at the aspect-level.
Opinion mining (also called sentiment analysis) is a type of natural language processing for computing people's opinions and emotions. It detects opinions from structured, semi-structured, and unstructured social media contents at different levels, such as the document, word, sentence, and aspect levels. In all these levels except aspect, opinion mining identifies the overall subjectivity or sentiment polarities. An aspect level is described as a part or an attribute of an entity. It exactly describes people's likes and dislikes in social media contents. In this paper, we propose a new framework for ranking products based on aspects. First, the system identifies the aspects of products. Second, the aspects and their opinion words are identified and visualized from the products’ reviews using a Harel–Koren fast multiscale layout. Third, the network visualization is constructed and modeled, and a Spearman's rank correlation coefficient based opinion ranking method is applied to rank the products based on positive and negative ranks. Fourth, the supervised learning methods (Naïve Bayes, Maximum Entropy, and Support Vector Machine) are employed for the aspect-based sentiment classification task. Finally, the performance of the system is measured by the experimental results.
Fusing texture, shape and deep model-learned information at decision level for automated classification of lung nodules on chest CT
2018, Information Fusion
Citation Excerpt :
Radiologists globally typically visually analyze chest CT scans on a slice-by-slice basis, which is time-consuming, expensive and prone to reader bias and requires a high degree of skill and concentration. Computer-aided diagnosis (CAD), however, avoids many of these issues and is increasingly being investigated as an alternative and complementary approach to conventional reading [4]. Many automated lung nodule classification approaches have been proposed in the literature and most of them consist of image preprocessing, nodule detection, nodule segmentation, feature extraction and classification.
The separation of malignant from benign lung nodules on chest computed tomography (CT) is important for the early detection of lung cancer, since early detection and management offer the best chance for cure. Although deep learning methods have recently produced a marked improvement in image classification there are still challenges as these methods contain myriad parameters and require large-scale training sets that are not usually available for most routine medical imaging studies. In this paper, we propose an algorithm for lung nodule classification that fuses the texture, shape and deep model-learned information (Fuse-TSD) at the decision level. This algorithm employs a gray level co-occurrence matrix (GLCM)-based texture descriptor, a Fourier shape descriptor to characterize the heterogeneity of nodules and a deep convolutional neural network (DCNN) to automatically learn the feature representation of nodules on a slice-by-slice basis. It trains an AdaBoosted back propagation neural network (BPNN) using each feature type and fuses the decisions made by three classifiers to differentiate nodules. We evaluated this algorithm against three approaches on the LIDC-IDRI dataset. When the nodules with a composite malignancy rate 3 were discarded, regarded as benign or regarded as malignant, our Fuse-TSD algorithm achieved an AUC of 96.65%, 94.45% and 81.24%, respectively, which was substantially higher than the AUC obtained by other approaches.
Psoriasis skin biopsy image segmentation using Deep Convolutional Neural Network
2018, Computer Methods and Programs in Biomedicine
Citation Excerpt :
Fig. 1 illustrates a pictorial representation of normal and psoriasis affected human skin biopsy images. Now a days, computer vision and image processing techniques are widely used in diverse areas in medical image analysis for automatic diagnosis [10–13], severity assessment [14–16]. Application of image processing techniques for automatic tool development certainly reduces the work load of the pathologists, standardizes the analysis (reduces inter-observer variations) and is able to extract accurate quantitative information in an inexpensive manner.
Development of machine assisted tools for automatic analysis of psoriasis skin biopsy image plays an important role in clinical assistance. Development of automatic approach for accurate segmentation of psoriasis skin biopsy image is the initial prerequisite for developing such system. However, the complex cellular structure, presence of imaging artifacts, uneven staining variation make the task challenging. This paper presents a pioneering attempt for automatic segmentation of psoriasis skin biopsy images.
Several deep neural architectures are tried for segmenting psoriasis skin biopsy images. Deep models are used for classifying the super-pixels generated by Simple Linear Iterative Clustering (SLIC) and the segmentation performance of these architectures is compared with the traditional hand-crafted feature based classifiers built on popularly used classifiers like K-Nearest Neighbor (KNN), Support Vector Machine (SVM) and Random Forest (RF). A U-shaped Fully Convolutional Neural Network (FCN) is also used in an end to end learning fashion where input is the original color image and the output is the segmentation class map for the skin layers.
An annotated real psoriasis skin biopsy image data set of ninety (90) images is developed and used for this research. The segmentation performance is evaluated with two metrics namely, Jaccard’s Coefficient (JC) and the Ratio of Correct Pixel Classification (RCPC) accuracy. The experimental results show that the CNN based approaches outperform the traditional hand-crafted feature based classification approaches.
The present research shows that practical system can be developed for machine assisted analysis of psoriasis disease.

View all citing articles on Scopus

Wei Huang obtained his B.Eng. and M.Eng. degrees from Harbin Institute of Technology, China, in 2004 and 2006, respectively. He obtained his Ph.D. degree from Nanyang Technological University, Singapore, in 2011. Before joining Nanchang University as Associate Professor, he worked in University of California San Diego, USA, and Agency for Science Technology and Research, Singapore, as Research Associate and Research Fellow, respectively. Dr. Huang has published 40+ academic papers and won the best paper award of MICCAI-MLMI in 2010. He is also the principal investigator in 2 NSFC grants and several other national/provincial grants at present. Dr. Huang׳s research interests mainly include but not limited to medical image processing, pattern recognition, and computer vision.

Shuru Zeng obtained her B.Eng. degree from Jiangxi Agricultural University in 2014. She is now a M.Eng. candidate in Nanchang University under the supervision of Prof. Wei Huang. Her research interests mainly include image processing, computer vision and pattern recognition.

Min Wan received B.Eng. and M.Eng. degrees from Beijing University of Post and Telecom as well as China Academy of Telecom Technology, Beijing, China, in 2004 and 2008, respectively, and was awarded the Ph.D. degree by Nanyang Technological University in 2012. He worked in National Heart Centre Singapore as Research Fellow from 2012 to 2013, and then a scientist in Institute of High Performance Computing, Agency for Science, Technology and Research (A⁎STAR) from 2013 to 2014. He is now an Associate Professor in Nanchang University. His research interests include medical imaging, computational geometry, computer graphics, and mesh generation.

Guang Chen received B.E. degree from the Nanjing Institute of Communication Engineering, China, in 2001. He received his M.E. degree from the Xidian University, China, in 2008. He is now a faculty in Xi׳an Communications Institute. His current research interests include software engineering, information forensics and signal processing.

View full text