Maitreya Maity

A Web-Accessible Framework for Automated Storage with Compression and Textural Classification of Malaria Parasite Images

Malaria being one of the serious health burdens especially in Indian population is conventionally diagnosed by expert pathologists through microscopic observation of stained peripheral blood smears. In order to provide rapid and efficient healthcare support to the common people at rural areas where experts are not (often) available, there is indeed a requirement of developing web-enabled healthcare system. In view of this, in this study, a web-accessible framework for automated storage of compressed microscopic images and texture-based screening of malaria parasite has been developed to provide rapid and efficient diagnosis even at remote public health clinics. It consists of (a) automated storage of microscopic images followed by JPEG image compression for faster transmission; (b) watershed transform based erythrocyte segmentation followed by image preprocessing; (c) texture feature extraction and selection; and (d) supervised classification and validation. Here, total 76 textures are extracted from segmented erythrocytes. Twenty six significant features are selected by using SVM based recursive feature elimination (SVM-RFE) method. Thereafter, supervised classifiers viz. Naive Baye’s approach, C4.5 and NBTree are considered for six-class classification problem and their performance are compared. From the result, it has been found that NBTRee classifier provides higher accuracy to classify P. vivax and P. falciparum (sensitivity: 99.0%, specificity: 99.8%) with different stages viz. ring, gametocytes and scizon under our developed web-accessible framework.

Computer-assisted approach to anemic erythrocyte classification using blood pathological information

Pathological blood test is one of the most important key issues in medical field prior to disease diagnosis. The aim of this paper is to design and develop a standalone application for the purpose of both acquisition and management of patient blood pathological information and generate automated anemia diagnosis report using computer vision approach. The developed system can be deployed in any pathological laboratory to help pathologist by giving support of automated anemia diagnosis and computerized report generation. Advanced image processing algorithm and data mining approach have been used to analysis patient medical information. The pathological data analysis module can process the blood test result to detect anemia type in blood. The image analysis module can identify the abnormal erythrocytes in the smear images using shape based classification. A total number of 38 shape features are extracted from each erythrocyte. Moreover, the supervised decision tree classifier C4.5 is used to classify image samples with sensitivity of 98.1% and specificity of 99.6%. The proposed system will record patient medical information like clinical data, blood test data, and microscopic smear images. Java swing, ImageJ, Weka, Java cryptography extension etc. libraries have been used to develop different applications module of the proposed system.

Fuzzy spectral clustering for automated delineation of chronic wound region using digital images

Chronic wound is an abnormal disease condition of localized injury to the skin and its underlying tissues having physiological impaired healing response. Assessment and management of such wound is a significant burden on the healthcare system. Currently, precise wound bed estimation depends on the clinical judgment and remains a difficult task. The paper introduces a novel method for ulcer boundary demarcation and estimation, using optical images captured by a hand-held digital camera. The proposed approach involves gray based fuzzy similarity measure using spatial knowledge of an image. The fuzzy measure is used to construct similarity matrix. The best color channel was chosen by calculating the mean contrast for 26 different color channels of 14 color spaces. It was found that Db color channel has highest mean contrast which provide best segmentation result in comparison with other color channels. The fuzzy spectral clustering (FSC) method was applied on Db color channel for effective delineation of wound region. The segmented wound regions were effectively post-processed using various morphological operations. The performance of proposed segmentation technique was validated by ground-truth images labeled by two experienced dermatologists and a surgeon. The FSC approach was tested on 70 images. FSC effectively segmented targeted ulcer boundary yielding 91.5% segmentation accuracy, 86.7%, Dice index and 79.0%. Jaccard score. The sensitivity and specificity was found to be 87.3% and 95.7% respectively. The performance evaluation shows the robustness of the proposed method of wound area segmentation and its potential to be used for designing patient comfort centric wound care system.

An Ensemble Rule Learning Approach for Automated Morphological Classification of Erythrocytes

The analysis of pathophysiological change to erythrocytes is important for early diagnosis of anaemia. The manual assessment of pathology slides is time-consuming and complicated regarding various types of cell identification. This paper proposes an ensemble rule-based decision-making approach for morphological classification of erythrocytes. Firstly, the digital microscopic blood smear images are pre-processed for removal of spurious regions followed by colour normalisation and thresholding. The erythrocytes are segmented from background image using the watershed algorithm. The shape features are then extracted from the segmented image to detect shape abnormality present in microscopic blood smear images. The decision about the abnormality is taken using proposed multiple rule-based expert systems. The deciding factor is majority ensemble voting for abnormally shaped erythrocytes. Here, shape-based features are considered for nine different types of abnormal erythrocytes including normal erythrocytes. Further, the adaptive boosting algorithm is used to generate multiple decision tree models where each model tree generates an individual rule set. The supervised classification method is followed to generate rules using a C4.5 decision tree. The proposed ensemble approach is precise in detecting eight types of abnormal erythrocytes with an overall accuracy of 97.81% and weighted sensitivity of 97.33%, weighted specificity of 99.7%, and weighted precision of 98%. This approach shows the robustness of proposed strategy for erythrocytes classification into abnormal and normal class. The article also clarifies its latent quality to be incorporated in point of care technology solution targeting a rapid clinical assistance.

Pixel-Based Supervised Tissue Classification of Chronic Wound Images with Deep Autoencoder

With the extensive use of machine vision methodologies, computer-assisted disease diagnosis has become a popular practice for the medical professionals. Detailed analysis of wound bed area and precise identification of the wound tissue regions are the most desirable aspects of an automated wound assessment applications. This study proposes a supervised wound tissue classification method, where a deep neural network classifier model is trained by the colour, texture and statistical features which are extracted from different tissue regions. The proposed classification process considers three types of tissue, viz. granulation (red), necrotic (black) and slough (yellow) and a total of 105 features are used for the classification. A pixel-based feature extraction approach is implemented to extract features from the tissue region, where a mask window of size \(9\,\times \,9\) runs over each pixel of the tissue regions for feature extraction. The proposed deep neural network achieves accuracy 99.997215%, sensitivity 99.998006%, specificity 99.996625% and F-Measure 99.997316%.

Selection of Colour Correction Algorithms for Calibrating Optical Chronic Ulcer Images

Visual observation is the gold standard for chronic skin ulcer assessment. The colour is the primary parameter for the visual evaluation and essential feature for automated wound diagnosis and monitoring. Inconsistent camera configuration and lighting condition can make image acquisition most difficult. Such external effects cause changes in the colour of digital images. This paper attempts to find a colour correction method that can calibrate optical ulcer images in respect of any practical conditions. Here, total 12 colour correction methods are compared using ten image quality metrics to find the best solution. A 24-bit Macbeth colour checker was also used to generate a reference image for the comparison. Based on performance analysis results, Weighted Grey Edge methods can correct wound images most accurately than other correction methods.

Combining GMM-Based Hidden Markov Random Field and Bag-of-Words Trained Classifier for Lung Cancer Detection Using Pap-Stained Microscopic Images

Lung cancer is a malignant tumour having uncontrolled lung cell growth. Papanicolaou (Pap)-stained cell cytology from Fine Needle Aspiration Cytology (FNAC) is the most followed approach for lung cancer diagnosis. However, the manual assessment of cytopathology slides under light microscopy is time consuming and suffers from feature ambiguities including inter-observer variability. Here, an automated computer vision approach is presented for identifying and classifying cancerous cell nuclei from pap-stained microscopic image of lung FNAC sample. The proposed methodology adopted Gaussian mixture model-based hidden Markov random field technique to segment cell nucleus. Later, bag-of-visual words model was used for nucleus classification, where scale-invariant feature transform feature were extracted from segmented nucleus for training a random forest classifier model. The adopted nucleus segmentation-cum-classification model was able to precisely segment the nucleus and classify them in two class, viz. Small Cell Lung Cancer (SCLC) and Non-small Cell Lung Cancer (NSCLC). The segmentation process achieves a sensitivity of 98.88% and specificity of 97.93%. And also, the nucleus classification model was able to perform with a sensitivity of 97.31%, specificity of 99.54%, and accuracy of 98.78%.

Assessment of Segmentation Techniques for Chronic Wound Surface Area Detection

A skin ulcer is a clinical pathology of localized damage to skin and tissue instigated by venous insufficiency. Precise identification of wound surface area is one of the challenging tasks in the dermatological evaluation. The assessment is carried out by clinicians using traditional approach of scales or metrics through visual inspection. The manual assessment leads to intra-observer variability, subjective error and time complexity. This paper evaluates the performances of supervised and unsupervised segmentation techniques used for wound area detection. The unsupervised methods used for evaluation were namely K-means, Fuzzy C-means and Gaussian mixture model. On the other part, random forest was implemented for supervised classification. Several filtering methods were used to generate image feature set from wound images to train random forest. The Gaussian mixture model with classification expectation–maximization clustering method achieved the highest weighted sensitivity of 95.91% and weighted specificity of 96.7%. The comparative study shows the superiority of proposed method and its suitability in wound segmentation from normal skin.

A Computer Vision Approach for Lung Cancer Classification Using FNAC-Based Cytological Images

Lung cancer represents malignant tumour having uncontrolled lung cell growth/proliferation. It can be diagnosed by invasive and non-invasive diagnostic approaches. One of the most effective and accurate approach is Papanicolaou (Pap)-stained cell cytology from fine needle aspiration cytology (FNAC). The manual assessment of cytopathology slides under light microscopy is time-consuming and suffers from feature ambiguities including inter-observer variability. To overcome such problems, the automated cytological analysis is the need of time. This study presents an automated computer vision approach to identify and classify cancerous cell present in microscopic images of Pap smear. The proposed methodology follows colour normalization, image filtering, nucleus segmentation and classification of segmented cells. The nucleus is segmented using the Random Walker with K-means clustering method. The post-processing is carried out on the segmented images to delineate joined nucleus and to remove unwanted regions. Subsequently, multiple nuclear features, i.e. colour, texture and geometric attributes are extracted from each segmented nucleus. After that, a comparative study on supervised classifier selection for the extracted features was adopted towards improving classification accuracy for distinguishing nucleus of non-small cell and small cell lung cancer. Artificial neural network performs best with sensitivity of \( 97.58\% \), specificity of \( 97.6\% \), accuracy of \( 97.46\% \).

Web-Enabled Distributed Health-Care Framework for Automated Malaria Parasite Classification: an E-Health Approach

Web-enabled e-healthcare system or computer assisted disease diagnosis has a potential to improve the quality and service of conventional healthcare delivery approach. The article describes the design and development of a web-based distributed healthcare management system for medical information and quantitative evaluation of microscopic images using machine learning approach for malaria. In the proposed study, all the health-care centres are connected in a distributed computer network. Each peripheral centre manages its’ own health-care service independently and communicates with the central server for remote assistance. The proposed methodology for automated evaluation of parasites includes pre-processing of blood smear microscopic images followed by erythrocytes segmentation. To differentiate between different parasites; a total of 138 quantitative features characterising colour, morphology, and texture are extracted from segmented erythrocytes. An integrated pattern classification framework is designed where four feature selection methods viz. Correlation-based Feature Selection (CFS), Chi-square, Information Gain, and RELIEF are employed with three different classifiers i.e. Naive Bayes’, C4.5, and Instance-Based Learning (IB1) individually. Optimal features subset with the best classifier is selected for achieving maximum diagnostic precision. It is seen that the proposed method achieved with 99.2% sensitivity and 99.6% specificity by combining CFS and C4.5 in comparison with other methods. Moreover, the web-based tool is entirely designed using open standards like Java for a web application, ImageJ for image processing, and WEKA for data mining considering its feasibility in rural places with minimal health care facilities.