Publications

Google Scholar Citations

To receive notifications about my future publications visit here

Automated Detection of Lung Nodules Using HOG Technique with Chest X-Ray Images (Published)

Lung disease is a growing disease and hence needs a lot of attention. It is challenging to delineate the boundary of the lung when it is imaged through an X-ray due to poor resolution. Hence, computer-aided diagnosis (CAD) is preferred as it assists the radiologists' inefficient diagnosis. In this work, a novel supervised classification technique is proposed using the histogram of oriented gradient (HOG) and neighborhood preserving embedding (NPE). Our method is evaluated using 2000 chest X-ray images and can efficiently classify normal and abnormal classes with a promising performance of 97.95% accuracy, using a support vector machine (SVM) classifier.

The Paper can be found here

A Two-Layer Sparse Autoencoder for Glaucoma Identification with Fundus Images (Published)

Glaucoma is a type of eye condition which may result in partial or consummate vision loss. Higher intraocular pressure is the leading cause of this condition. Screening for glaucoma and early detection can avert vision loss. Computer-aided diagnosis (CAD) is an automated process with the potential to identify glaucoma early through quantitative analysis of digital fundus images. Preparing an effective model for CAD requires a large database. This study presents a CAD tool for the precise detection of glaucoma using a machine learning approach. An autoencoder is trained to determine effective and important features from fundus images. These features are used to develop classes of glaucoma for testing. The method achieved an F − measure value of 0.95 utilizing 1426 digital fundus images (589 control and 837 glaucomas). The efficacy of the system is evident and is suggestive of its possible utility as an additional tool for verification of clinical decisions

The paper has been published in the Springer journal of medical systems and can be found here

Computer-aided diagnosis for the identification of breast cancer using thermogram images: A comprehensive review (Published)

Breast cancer is cancer that can form in the cells of breasts. It is much more common in females than in males. The most common periods of cancer development are during puberty, pregnancy, and breastfeeding. Thermography can be utilized for breast analysis and provides useful data on the location of hyperthermia and the vascular state of the tissue. Computer-Aided diagnosis is an algorithmic approach that can be assistive during routine screening so that human error in breast analysis for cancer detection is reduced. In early-stage cancer, the accuracy of the assessment then increases, enabling clinicians to make an improved diagnosis of benign versus malignant classification. Herein, we have reviewed thermogram-based computer-aided diagnostic systems developed during the last two decades for breast cancer screening and analysis. We explore the quantitative and qualitative performances of machine learning-based approaches, which include segmentation-based and feature extraction based methods, dimensionality reduction, and various classification schemes, as proposed in the literature. We also describe the limitations, as well as future requirements to improve current techniques, which can help researchers and clinicians to be apprised of quantitative developments and to plan for the future.

The paper has been published in the springer Infrared physics and technology journal and can be found here.

Feature versus Deep Learning Based Approaches for the Automated Detection of Brain Tumor with MR Images: A Comparative Study (Published)

The public health is significantly affected by development of brain tumors in human patients. Because of limited efficacy so far, this affliction is of great interest for devising better quantitative detection and treatment methods. Glioblastoma (GBM) is a relatively common, malignant form of brain tumor, which is currently challenging to treat and cure. In contrast, lower grade gliomas (LGG) originate from glial cells and can mostly be treated and cured in the initial stages if they are detected. Yet, detection and assessment of both diseases at an early stage is difficult. A computer-aided diagnostic (CAD) tool may help to test for the presence and extent of any such tumor, and thus can be assistive in the clinical diagnostic process. In this paper, we have compared handcrafted versus non-handcrafted features based new CAD systems to characterize GBM and LGG. Our machine learning based handcrafted model uses quantitative techniques of enhanced elongated quinary patterns and entropies analysis. The entropies are employed to extract prominent features from brain magnetic resonance images acquired from suspected patients. Classification of the imagery is done using a ranked features method. We have also developed a nonhandcrafted based deep learning model using VGG - 16 architecture for segregating GBM and LGG subjects. Our study yielded a maximum classification accuracy of 94.25% for non-handcrafted features with ten-fold cross-validation using the k-nearest neighbor classifier. Thus, our results suggest the efficacy of the proposed CAD model for assisting in the diagnostic process

The Paper is Published in the International Journal of Imaging Systems and Technology and can be found here

Elongated Quinary Patterns based Entropies (EQPE) for the Automated Detection of Brain Tumor with MR Images (Paper under review)

Public health is significantly affected by the development of brain tumors in human patients. Because of limited efficacies so far, this affliction is of great interest in devising better quantitative detection and treatment methods. Glioblastoma (GBM) is a relatively common, malignant form of brain tumor, which is currently challenging to treat and cure. In contrast, lower-grade gliomas (LGG) originate from glial cells and can mostly be treated and cured in the initial stages if they are detected. Yet, detection and assessment of both diseases at an early stage is difficult. A computer-aided diagnostic (CAD) tool may help to test for the presence and extent of any such tumor, and thus can be assistive in the clinical diagnostic process. In this paper, we propose a new CAD system to characterize GBM and LGG using the quantitative techniques of enhanced elongated quinary patterns and entropies analysis. The entropies are employed to extract prominent features from brain magnetic resonance (MR) images acquired from suspected patients. Classification of the imagery is done using a ranked features method. Our study yielded a maximum classification accuracy of 92.5% with ten-fold cross-validation using the k-nearest-neighbor classifier. Thus our results suggest the efficacy of the proposed CAD model for assisting in the diagnostic process.

This paper is under review in the IEEE Journal of Computational Biology and Bioinformatics

Automated detection of Hypertrophic cardiomyopathy using Emperical wavelet transformation and higher order-spectra techniques with Echocardiography Images (Paper under review)

Hypertrophic cardiomyopathy (HCM) is a genetic disorder that exhibits a wide spectrum of clinical presentations ranging from absence of any symptom to sudden cardiac death (SCD) due to malignant arrhythmia. HCM is characterized by a nondilated thick-walled ventricle with altered diastolic function and preserved systolic function at the initial phase of the disease course. Extensive hypertrophy in the interventricular septum may lead to left ventricular outflow tract obstruction (LVOTO). In addition, the thick myocardium in HCM may lead to ischemia, resulting in chest discomfort, and may also contribute to progressive heart failure with dyspnea and related symptoms, and sudden cardiac death from ventricular tachyarrhythmia. The management of HCM patients depends on the extent of hypertrophy, the presence of LVOTO, arrhythmia, and symptoms. The different therapies focus on surrogate outcomes, and there is no definitive cure for the disease to date. The gold standard in the diagnosis of HCM is the identification of the genetic defect causing it, of which there are several. If genetic testing was not available, experienced sonographers can make the diagnosis using echocardiography with the background clinical information. However, this process is operator-dependent and subject to human error. Therefore, we propose a Computer-Aided Diagnosis (CAD) method to discriminate HCM from healthy subjects. We used 113 echocardiography images for our study. These were resized to 256 × 256 resolution. For preprocessing, we used empirical wavelet transforms, because the technique generates a set of filter banks based on the component frequencies of the image, which are useful to identify image boundaries. We then derived eight Littlewood-Paley coefficients from the images. From these eight coefficients, 72 higher order spectra (HOS) features were extracted. In total, we extracted 576 (72×8) features from a single image. These features were ranked using the Wilcoxon ranking test, after which the ranked features were input to a support vector machine (SVM) classifier with polynomial and random basis function (RBF).

The highest classification accuracy of 80.83% was obtained using the SVM-RBF classifier.

This paper is under review in the Elsevier journal of Ultrasonics

Detection of fetal myocardial hypertrophy using multi-resolution techniques: An early detection approach (Paper Under Review)

In the presence of gestational diabetes mellitus (GDM), the fetus is subjected to a hyperinsulinemia environment. This environment can cause a wide range of metabolic and fetal cardiac structural alterations. Fetal myocardial hypertrophy predominantly affecting the interventricular septum was found to have a morphology of disarray similar to hypertrophic cardiomyopathy (HCM) and was observed to be present in some GDM neonates after birth. Myocardial thickness might increase in GDM fetuses irrespective of glycemic control status and fetal weight. Fetal echocardiography performed in fetuses of GDM helps in assessing cardiac structure and function, and to diagnose myocardial hypertrophy. Although ultrasonography eases the visualization of cardiac structure and thickness evaluation, there are no studies in the literature which have established evidence for morphologic variation associated with cardiac hypertrophy among fetuses of GDM mothers when assessed non-invasively. In this study, normal, pregestational diabetes mellitus (pre GDM) and GDM ultrasound images are initially pre-processed and subjected to multi-resolution analysis using the Gabor wavelet, empirical wavelet transform, and the discrete wavelet transform. Entropy features are extracted from the obtained coefficients, and data dimensionality reduction techniques in the form of Locality Sensitive Discriminant Analysis and Neighborhood Preserving Embedding were utilized for further processing. The resultant features were ranked and tested with different classifiers. Maximum accuracy of 80.39% was obtained with the combination of Gabor wavelet with entropy features for support vector machine with radial basis function kernel. This paradigm can be helpful to physicians for the differentiation of benign versus pathological cases.

This paper is under review in the Elseiver journal of Biocybernetics and Biomedical Engineering