Immanuel Alex Pandian

Bio: Immanuel Alex Pandian is an academic researcher from Karunya University. The author has contributed to research in topics: Motion estimation & Binary search algorithm. The author has an hindex of 2, co-authored 4 publications receiving 5 citations.

Early detection of Alzheimer’s disease using ensemble of pre-trained models

Abstract: Alzheimer’s disease is a neurological state in which the damage of neurons causes memory disabilities. Automatic early detection of Alzheimer’s disease (AD) is always demanding the competence of a machine to distinguish different dementia stages. The different dementia stages related to the human cognitive state are cognitively normal (CN), mild cognitive impairment convertible (MCIc), mild cognitive impairment nonconvertible (MCInc), and Alzheimer’s disease (AD). Both the stages MCIc and MCInc show the similar symptoms of memory disorders. The stage MCInc is quite normal in all aged people. The memory problems are due to normal aging. But the stage MCIc will lead to AD in few years. The stage MCIc is the early stage of AD. Proper meditation can reduce the rate of memory loss if the disease is identified in the early stage MCIc. The two dementia stages included in this study are cognitively normal and mild cognitive impairment convertible. In this work, an ensemble model is used to improve the classification accuracy between CN and MCIc. The ensemble model consists of two pre-trained network models Xception and MobileNet. The performance of pre-trained and ensemble models is tested using Alzheimer’s disease neuroimaging initiative (ADNI) dataset. The classification between CN and MCIc is crucial for the early detection of AD. The classification accuracy between MCIc vs CN is only included in this study. The accuracy obtained using Xception and MobileNet models are 89.23% and 89.89%. The proposed model provides a classification accuracy of 91.3%. The results show that the proposed ensemble model is providing good discrimination capability of the stages MCIc and CN.

Classification of Lung Nodule Using Hybridized Deep Feature Technique

Abstract: Deep learning techniques have become very popular among Artificial Intelligence (AI) techniques in many areas of life. Among many types of deep learning techniques, Convolutional Neural Networks (CNN) can be useful in image classification applications. In this work, a hybridized approach has been followed to classify lung nodule as benign or malignant. This will help in early detection of lung cancer and help in the life expectancy of lung cancer patients thereby reducing the mortality rate by this deadly disease scourging the world. The hybridization has been carried out between handcrafted features and deep features. The machine learning algorithms such as SVM and Logistic Regression have been used to classify the nodules based on the features. The dimensionality reduction technique, Principle Component Analysis (PCA) has been introduced to improve the performance of hybridized features with SVM. The experiments have been carried out with 14 different methods. It has been found that GLCM + VGG19 + PCA + SVM outperformed all other models with an accuracy of 94.93%, sensitivity of 90.9%, specificity of 97.36% and precision of 95.44%. The F1 score was found to be 0.93 and the AUC was 0.9843. The False Positive Rate was found to be 2.637% and False Negative Rate was 9.09%.

An efficient motion estimation algorithm based on particle swarm optimization

Abstract: The PSO algorithm reduce the search points without the degradation of the image quality. It provides accurate motion estimation with very low complexity in the context of video estimation. This algorithm is capable of reducing the computational complexity of block matching process. This algorithm maintains high estimation accuracy compared to the full search method. The critical component in most block-based video compression system is Motion Estimation because redundancy between successive frames of video sequence allows for compression of video data. These algorithms are used to reduce the computational requirement by checking only some points inside the search window, while keeping a good error performance when compared with Full Search and Diamond search algorithm. This algorithm should maintain high estimation accuracy compared to the Full search method and Diamond search algorithm. Here by using the PSO algorithm could get a high accuracy in the block-based motion estimation.

Review on Local Feature Descriptors for Early Detection of Alzheimer’s Disease

Abstract: Alzheimer’s disease (AD) is a brain disorder that affects the nerve cells. Gradually the patient is not able to do the daily routines and activities. Medical experts can not completely cure this disease. But if this disease is diagnosing as much as early, then they can start some medicines to reduce rate of brain damage. Also, the patient and his relatives can be prepared themselves. Hence early detection of Alzheimer’s disease is one of the major problems faced by the medical field. The four stages of Cognitive state of a human brain can be classified as Alzheimer’s disease (AD), Mild Cognitive Impairment Convertible (MCIc), Mild Cognitive Impairment Non-Convertible (MCInc) and Cognitive Normal (CN). MCIc is a stage which can be lead to AD in few years. MCInc is a stage in which cognitive abnormalities will be present, but it will not lead to AD. In the context of early detection of AD, MCIc vs MCInc classification is decisive. In this work, feature descriptors used for the classification of above four stages were studied with MRI Images. We emphasized the performance of different local feature descriptors. After the review, concluded that classification accuracy between MCIc vs MCInc should be improved for the early detection of AD

Ensemble learning approach for multi-class classification of Alzheimer’s stages using magnetic resonance imaging

Abstract: Alzheimer’s disease (AD) is a gradually progressing neurodegenerative irreversible disorder. Mild cognitive impairment convertible (MCIc) is the clinical forerunner of AD. Precise diagnosis of MCIc is essential for effective treatments to reduce the progressing rate of the disease. The other cognitive states included in this study are mild cognitive impairment non-convertible (MCInc) and cognitively normal (CN). MCInc is a stage in which aged people suffer from memory problems, but the stage will not lead to AD. The classification between MCIc and MCInc is crucial for the early detection of AD. In this work, an algorithm is proposed which concatenates the output layers of Xception, InceptionV3, and MobileNet pre-trained models. The algorithm is tested on the baseline T1-weighted structural magnetic resonance imaging (MRI) images obtained from Alzheimer’s disease neuroimaging initiative database. The proposed algorithm provided multi-class classification accuracy of 85%. Also, the proposed algorithm gave an accuracy of 85% for classifying MCIc vs MCInc, an accuracy of 94% for classifying AD vs CN, and an accuracy of 92% for classifying MCIc vs CN. The results exhibit that the proposed algorithm outruns other state-of-the-art methods for the multi-class classification and classification between MCIc and MCInc.

Comparative Study of Multiple Feature Descriptors for Detecting the Presence of Alzheimer’s Disease

Abstract: Medical image processing has a very important role in medical diagnosis where a doctor can compare the scanned image of his patient with a heap of images and find the result of the image that matches with it. With the help of feature descriptors, we can make the process of image classification much more efficient. By implementing various feature descriptors, we are able to identify Alzheimer’s at the very early stages which helps the entire curing process faster. This paper presents the comparison of various binary descriptors such as local binary pattern (LBP), local wavelet pattern (LWP), histogram-oriented gradients (HOG), local bit plane decoded pattern (LBDP) along with K-nearest neighbour (KNN) for its classification. The results indicate that the combination of LBP and KNN together produce a better accuracy of 91.21% in “Alzheimer’s Dataset” ( Alzheimer's Dataset (4 class of Images) https://www.kaggle.com/tourist55/alzheimers-dataset-4-class-of-images [1]) when compared to other descriptors.

Early detection of Alzheimer’s disease using local binary pattern and convolutional neural network

Abstract: Alzheimer’s disease, a progressive and irreversible abnormality of the human brain impairs memory and thinking skills. Gradually, it will damage the ability to carry out simple tasks. Even though the disease cannot be completely cured by medical specialists, the rate of brain damage can be pared if the disease is identified in its budding stage itself. Thus, victims and their relatives will get ample time to prepare themselves. Alzheimer’s disease (AD), cognitively normal (CN), mild cognitive impairment convertible (MCIc), and mild cognitive impairment non-convertible (MCInc) are the different phases of cognition. The state of memory loss in aged people, which will not lead to AD, can be encountered as MCInc. The state-MCIc gradually leads to AD. The work is intended for the early detection of AD. Early detection can be claimed if and only if the state-MCIc is detected. But the clinical visual identification of state-MCIc from MRI scan is difficult. In this work, a novel local feature descriptor is proposed for the detection of state-MCIc. The proposed local feature descriptor combined strengths of fast Hessian detector and local binary pattern texture operator for the identification of key points and descriptions. A simple convolutional neural network is used for classification. The classification accuracy between MCIc and CN is obtained as 88.46% which is a pivotal result for early detection of AD. The classification accuracy between AD and CN is attained at 88.99%. The results indicate that the proposed system can contribute a colossal innovation in the early detection of AD.

A Novel Hybridized Feature Extraction Approach for Lung Nodule Classification Based on Transfer Learning Technique

Abstract: Purpose: In the field of medical diagnosis, deep learning-based computer-aided detection of diseases will reduce the burden of physicians in the diagnosis of diseases especially in the case of lung cancer nodule classification. Materials and Methods: A hybridized model which integrates deep features from Residual Neural Network using transfer learning and handcrafted features from the histogram of oriented gradients feature descriptor is proposed to classify the lung nodules as benign or malignant. The intrinsic convolutional neural network (CNN) features have been incorporated and they can resolve the drawbacks of handcrafted features that do not completely reflect the specific characteristics of a nodule. In the meantime, they also reduce the need for a large-scale annotated dataset for CNNs. For classifying malignant nodules and benign nodules, radial basis function support vector machine is used. The proposed hybridized model is evaluated on the LIDC-IDRI dataset. Results: It has achieved an accuracy of 97.53%, sensitivity of 98.62%, specificity of 96.88%, precision of 95.04%, F1 score of 0.9679, false-positive rate of 3.117%, and false-negative rate of 1.38% and has been compared with other state of the art techniques. Conclusions: The performance of the proposed hybridized feature-based classification technique is better than the deep features-based classification technique in lung nodule classification.

Investigation of Deep Features in Lung Nodule Classification

Abstract: Cancer is one of the deadliest diseases that affect people worldwide, regardless of their socioeconomic situation. Head and neck cancer, brain cancer, stomach cancer, breast cancer, and pancreatic cancer are some of the different types of cancer. Lung cancer has the highest occurrence and fatality rate among all cancer diseases worldwide. It has become more prevalent in developing countries due to increased air pollution. As a result of this terrible sight, lung cancer screening and early diagnosis are now more important than ever. Taking Computed Tomography (CT) images of the entire chest region and analyzing them for any abnormalities is the most common method for detecting lung cancer at its initial stage. In the clinical diagnosis of many diseases, transfer learning and deep learning are becoming increasingly important. Lung nodules, categorized as malignant or benign, are radiographic indications of lung cancer. The application of transfer learning and deep learning in the Computed Aided Diagnosis (CAD) system to aid doctors in classifying lung nodules is proposed in this paper. Convolutional Neural Networks (CNNs) such as VGG16, VGG19, and ResNet50 are used as feature extractors in this study. In the classification of lung nodules, the performance of these deep features is explored.

Pattern Search Based on Particle Swarm Optimization Technique for Block Matching Motion Estimation Algorithm

Abstract: Block matching algorithm is a popular technique in developing video coding applications that is used to reduce the computational complexity of motion estimation (ME) algorithm. In a video encoder, efficient implementation of ME is required that affect the final result in any applications. Searching pattern is one of the factors in developing motion estimation algorithm that could provide good performance. A new enhanced algorithm using a pattern based particle swarm optimization (PSO) has been proposed for obtaining least number of computations and to give better estimation accuracy. Due to the center biased nature of the videos, the proposed algorithm approach uses an initial pattern to speed up the convergence of the algorithm. The results have proved that improvements over Hexagon base Search could achieved with 7.82%–17.57% of computations cost reduction without much value of degradation of image quality. This work could be improved by using other variant of PSO or other potential meta-heuristic algorithms to provide the better performances in both aspects.