Manish Chowdhury
@siemens.com
Siemens Corporate Research
EDUCATION
PhD (Image Processing and Machine learning) - Indian Statistical Institute Kolkata (Degree Awarded by University of Calcutta)
BE (Electronics)- University of Pune
Scopus Publications
Scholar Citations
Scholar h-index
Scholar i10-index
Scopus Publications
Lipi B. Mahanta, Elima Hussain, Navarun Das, Lopamudra Kakoti, and Manish Chowdhury
Elsevier BV
A. Sai Charan, M. Jitesh, M. Chowdhury, and H. Venkataraman
Institution of Engineering and Technology (IET)
Elima Hussain, Lipi B. Mahanta, Chandana Ray Das, Manjula Choudhury, and Manish Chowdhury
Elsevier BV
Pap smear is often employed as a screening test for diagnosing cervical pre-cancerous and cancerous lesions. Accurate identification of dysplastic changes amongst the cervical cells in a Pap smear image is thus essential for rapid diagnosis and prognosis. Manual pathological observations used in clinical practice require exhaustive analysis of thousands of cell nuclei in a whole slide image to visualize the dysplastic nuclear changes which make the process tedious and time-consuming. Automated nuclei segmentation and classification exist but are challenging to overcome issues like nuclear intra-class variability and clustered nuclei separation. To address such challenges, we put forward an application of instance segmentation and classification framework built on an Unet architecture by adding residual blocks, densely connected blocks and a fully convolutional layer as a bottleneck between encoder-decoder blocks for Pap smear images. The number of convolutional layers in the standard Unet has been replaced by densely connected blocks to ensure feature reuse-ability property while the introduction of residual blocks in the same attempts to converge the network more rapidly. The framework provides simultaneous nuclei instance segmentation and also predicts the type of nucleus class as belonging to normal and abnormal classes from the smear images. It works by assigning pixel-wise labels to individual nuclei in a whole slide image which enables identifying multiple nuclei belonging to the same or different class as individual distinct instances. Introduction of a joint loss function in the framework overcomes some trivial cell level issues on clustered nuclei separation. To increase the robustness of the overall framework, the proposed model is preceded with a stacked auto-encoder based shape representation learning model. The proposed model outperforms two state-of-the-art deep learning models Unet and Mask_RCNN with an average Zijdenbos similarity index of 97 % related to segmentation along with binary classification accuracy of 98.8 %. Experiments on hospital-based datasets using liquid-based cytology and conventional pap smear methods along with benchmark Herlev datasets proved the superiority of the proposed method than Unet and Mask_RCNN models in terms of the evaluation metrics under consideration.
M. Chowdhury, P. Das, and A. Das
Springer Science and Business Media LLC
This paper aims to investigate the global $$O(t^{-\\alpha })$$ synchronization of a class of fractional-order non-autonomous neural networks with time delay. Using centralized data-sampling principle and the theory of fractional differential equations, sufficient criteria for the $$O(t^{-\\alpha })$$ synchronization is derived. Centralized data-sampling control is applied in the drive-response-based coupled neural networks to achieve global $$O(t^{-\\alpha })$$ synchronization. It is a more effective strategy as it gives better control performance. Numerical examples are also given to illustrate the validity of the theoretical result.
M. Chowdhury and P. Das
Springer Singapore
We have considered a general form of non-autonomous cellular neural network with time varying delays in this paper. We have estimated the upper bound of solutions of the system by introducing different parameters and considered some conditions on it. We have derived the conditions of boundedness and global exponential stability of the model which is initially unstable for some parameter values using Young Inequality technique and Dini derivative. Several examples and their computer simulations are given to illustrate the effectiveness of obtained results.
Jaydeb Mondal, Malay Kumar Kundu, Sudeb Das, and Manish Chowdhury
Springer Science and Business Media LLC
The fundamental step in video content analysis is the temporal segmentation of video stream into shots, which is known as Shot Boundary Detection (SBD). The sudden transition from one shot to another is known as Abrupt Transition (AT), whereas if the transition occurs over several frames, it is called Gradual Transition (GT). A unified framework for the simultaneous detection of both AT and GT have been proposed in this article. The proposed method uses the multiscale geometric analysis of Non-Subsampled Contourlet Transform (NSCT) for feature extraction from the video frames. The dimension of the feature vectors generated using NSCT is reduced through principal component analysis to simultaneously achieve computational efficiency and performance improvement. Finally, cost efficient Least Squares Support Vector Machine (LS-SVM) classifier is used to classify the frames of a given video sequence based on the feature vectors into No-Transition (NT), AT and GT classes. A novel efficient method of training set generation is also proposed which not only reduces the training time but also improves the performance. The performance of the proposed technique is compared with several state-of-the-art SBD methods on TRECVID 2007 and TRECVID 2001 test data. The empirical results show the effectiveness of the proposed algorithm.
Amirreza Mahbod, Manish Chowdhury, Örjan Smedby, and Chunliang Wang
Elsevier BV
Segmenting brain tissue from MR scans is thought to be highly beneficial for brain abnormality diagnosis, prognosis monitoring, and treatment evaluation. Many automatic or semi-automatic methods ha ...
Manish Chowdhury and Malay Kumar Kundu
Springer Science and Business Media LLC
Malay Kumar Kundu, Manish Chowdhury, and Sudeb Das
Elsevier BV
BACKGROUND AND OBJECTIVE
Content based medical image retrieval (CBMIR) systems enable fast diagnosis through quantitative assessment of the visual information and is an active research topic over the past few decades. Most of the state-of-the-art CBMIR systems suffer from various problems: computationally expensive due to the usage of high dimensional feature vectors and complex classifier/clustering schemes. Inability to properly handle the "semantic gap" and the high intra-class versus inter-class variability problem of the medical image database (like radiographic image database). This yields an exigent demand for developing highly effective and computationally efficient retrieval system.
METHODS
We propose a novel interactive two-stage CBMIR system for diverse collection of medical radiographic images. Initially, Pulse Coupled Neural Network based shape features are used to find out the most probable (similar) image classes using a novel "similarity positional score" mechanism. This is followed by retrieval using Non-subsampled Contourlet Transform based texture features considering only the images of the pre-identified classes. Maximal information compression index is used for unsupervised feature selection to achieve better results. To reduce the semantic gap problem, the proposed system uses a novel fuzzy index based relevance feedback mechanism by incorporating subjectivity of human perception in an analytic manner.
RESULTS
Extensive experiments were carried out to evaluate the effectiveness of the proposed CBMIR system on a subset of Image Retrieval in Medical Applications (IRMA)-2009 database consisting of 10,902 labeled radiographic images of 57 different modalities. We obtained overall average precision of around 98% after only 2-3 iterations of relevance feedback mechanism. We assessed the results by comparisons with some of the state-of-the-art CBMIR systems for radiographic images.
CONCLUSIONS
Unlike most of the existing CBMIR systems, in the proposed two-stage hierarchical framework, main importance is given on constructing efficient and compact feature vector representation, search-space reduction and handling the "semantic gap" problem effectively, without compromising the retrieval performance. Experimental results and comparisons show that the proposed system performs efficiently in the radiographic medical image retrieval field.
Manish Chowdhury, Benjamin Klintström, Eva Klintström, Örjan Smedby, and Rodrigo Moreno
Springer International Publishing
The accuracy of the analyses for studying the three dimensional trabecular bone microstructure rely on the quality of the segmentation between trabecular bone and bone marrow. Such segmentation is challenging for images from computed tomography modalities that can be used in vivo due to their low contrast and resolution. For this purpose, we propose in this paper a granulometry-based segmentation method. In a first step, the trabecular thickness is estimated by using the granulometry in gray scale, which is generated by applying the opening morphological operation with ball-shaped structuring elements of different diameters. This process mimics the traditional sphere-fitting method used for estimating trabecular thickness in segmented images. The residual obtained after computing the granulometry is compared to the original gray scale value in order to obtain a measurement of how likely a voxel belongs to trabecular bone. A threshold is applied to obtain the final segmentation. Six histomorphometric parameters were computed on 14 segmented bone specimens imaged with cone-beam computed tomography (CBCT), considering micro-computed tomography (micro-CT) as the ground truth. Otsu’s thresholding and Automated Region Growing (ARG) segmentation methods were used for comparison. For three parameters (Tb.N, Tb.Th and BV/TV), the proposed segmentation algorithm yielded the highest correlations with micro-CT, while for the remaining three (Tb.Nd, Tb.Tm and Tb.Sp), its performance was comparable to ARG. The method also yielded the strongest average correlation (0.89). When Tb.Th was computed directly from the gray scale images, the correlation was superior to the binary-based methods. The results suggest that the proposed algorithm can be used for studying trabecular bone in vivo through CBCT.
Manish Chowdhury, Daniel Jörgens, Chunliang Wang, Örjan Smedby, and Rodrigo Moreno
SPIE
Cortical bone plays a big role in the mechanical competence of bone. The analysis of cortical bone requires accurate segmentation methods. Level set methods are usually in the state-of-the-art for segmenting medical images. However, traditional implementations of this method are computationally expensive. This drawback was recently tackled through the so-called coherent propagation extension of the classical algorithm which has decreased computation times dramatically. In this study, we assess the potential of this technique for segmenting cortical bone in interactive time in 3D images acquired through High Resolution peripheral Quantitative Computed Tomography (HR-pQCT). The obtained segmentations are used to estimate cortical thickness and cortical porosity of the investigated images. Cortical thickness and Cortical porosity is computed using sphere fitting and mathematical morphological operations respectively. Qualitative comparison between the segmentations of our proposed algorithm and a previously published approach on six images volumes reveals superior smoothness properties of the level set approach. While the proposed method yields similar results to previous approaches in regions where the boundary between trabecular and cortical bone is well defined, it yields more stable segmentations in challenging regions. This results in more stable estimation of parameters of cortical bone. The proposed technique takes few seconds to compute, which makes it suitable for clinical settings.
Kangkana Bora, Manish Chowdhury, Lipi B. Mahanta, Malay Kumar Kundu, and Anup Kumar Das
Elsevier BV
BACKGROUND AND OBJECTIVES
The present study proposes an intelligent system for automatic categorization of Pap smear images to detect cervical dysplasia, which has been an open problem ongoing for last five decades.
METHODS
The classification technique is based on shape, texture and color features. It classifies the cervical dysplasia into two-level (normal and abnormal) and three-level (Negative for Intraepithelial Lesion or Malignancy, Low-grade Squamous Intraepithelial Lesion and High-grade Squamous Intraepithelial Lesion) classes reflecting the established Bethesda system of classification used for diagnosis of cancerous or precancerous lesion of cervix. The system is evaluated on two generated databases obtained from two diagnostic centers, one containing 1610 single cervical cells and the other 1320 complete smear level images. The main objective of this database generation is to categorize the images according to the Bethesda system of classification both of which require lots of training and expertise. The system is also trained and tested on the benchmark Herlev University database which is publicly available. In this contribution a new segmentation technique has also been proposed for extracting shape features. Ripplet Type I transform, Histogram first order statistics and Gray Level Co-occurrence Matrix have been used for color and texture features respectively. To improve classification results, ensemble method is used, which integrates the decision of three classifiers. Assessments are performed using 5 fold cross validation.
RESULTS
Extended experiments reveal that the proposed system can successfully classify Pap smear images performing significantly better when compared with other existing methods.
CONCLUSION
This type of automated cancer classifier will be of particular help in early detection of cancer.
Kangkana Bora, Manish Chowdhury, Lipi B. Mahanta, Malay K. Kundu, and Anup K. Das
ACM Press
This article presents the result of a comprehensive study on deep learning based Computer Aided Diagnostic techniques for classification of cervical dysplasia using Pap smear images. All the experiments are performed on a real indigenous image database containing 1611 images, generated at two diagnostic centres. Focus is given on constructing an effective feature vector which can perform multiple level of representation of the features hidden in a Pap smear image. For this purpose Deep Convolutional Neural Network is used, followed by feature selection using an unsupervised technique with Maximal Information Compression Index as similarity measure. Finally performance of two classifiers namely Least Square Support Vector Machine (LSSVM) and Softmax Regression are monitored and classifier selection is performed based on five measures along with five fold cross validation technique. Output classes reflects the established Bethesda system of classification for identifying pre-cancerous and cancerous lesion of cervix. The proposed system is also compared with two existing conventional systems and also tested on a publicly available database. Experimental results and comparison shows that proposed system performs efficiently in Pap smear classification.
Manish Chowdhury, Samuel Rota Bulo, Rodrigo Moreno, Malay Kumar Kundu, and Orjan Smedby
IEEE
Content-Based Medical Image Retrieval (CBMIR) is an important research field in the context of medical data management. In this paper we propose a novel CBMIR system for the automatic retrieval of radiographic images. Our approach employs a Convolutional Neural Network (CNN) to obtain high-level image representations that enable a coarse retrieval of images that are in correspondence to a query image. The retrieved set of images is refined via a non-parametric estimation of putative classes for the query image, which are used to filter out potential outliers in favour of more relevant images belonging to those classes. The refined set of images is finally re-ranked using Edge Histogram Descriptor, i.e. a low-level edge-based image descriptor that allows to capture finer similarities between the retrieved set of images and the query image. To improve the computational efficiency of the system, we employ dimensionality reduction via Principal Component Analysis (PCA). Experiments were carried out to evaluate the effectiveness of the proposed system on medical data from the “Image Retrieval in Medical Applications” (IRMA) benchmark database. The obtained results show the effectiveness of the proposed CBMIR system in the field of medical image retrieval.
Manish Chowdhury and Malay Kumar Kundu
Springer Science and Business Media LLC
Recently, Content Based Image Retrieval (CBIR) has emerged as an active research area having applications in various fields. There exist several states-of-the art CBIR systems that uses both spatial and transform features as input. However, as hardly any details study reported so far on the effectiveness of different transform domain features in CBIR paradigm. This motivates the current article where we have presented extensive comparative assessment of five different transform domain features considering various filter combinations. Three different feature representation schemes and three different classifiers have been used for this purpose. Extensive experiments on four widely used benchmark image databases (Oliva, Caltech101, Caltech256 and MIRFlickr25000) were conducted to determine the best combination of transform, filters, feature representation and classifier. Furthermore, we have also attempted to discover the optimal features from the best combinations using maximal information compression index (MICI). Both qualitative and quantitative evaluations show that the combination of Least Square Support Vector Machine (LSSVM) as a classifier and the statistical parametric framework based reduced feature representation in Non-Subsampled Contourlet Transform (NSCT) with “pyrexc” and “sinc” filters gives the best retrieval performances.
Manish Chowdhury and Malay Kumar Kundu
ACM Press
We present a novel Content Based Medical Image Retrieval (CBMIR) scheme for color endoscopic images using Multi-scale Geometric Analysis (MGA) of Nonsubsampled Contourlet Transform (NSCT) and the statistical framework based on Generalized Gaussian Density (GGD) model and Kullback-Leibler Distance (KLD). The subband images obtained from the NSCT decomposition are divided into number of blocks and then the coefficients of each block of each subband is modeled with GGD parameters and computing the similarity using the KLD among the model parameters. The retrieval performance of the proposed system is further improved using Least Square-Support Vector Machine (LSSVM) classifier. Extensive experiments were carried out to evaluate the effectiveness of the proposed system on endoscopic image databases consisting of 276 images. Experimental results show that the proposed CBMIR system performs efficiently in image retrieval paradigm.
Malay Kumar Kundu, Manish Chowdhury, and Samuel Rota Bulò
Elsevier BV
Content-Based Image Retrieval (CBIR) is an important problem in the domain of digital data management. There is indeed a growing availability of images, but unfortunately the traditional metadata-based search systems are unable to properly exploit their visual information content. In this article we introduce a novel CBIR scheme that abstracts each image in the database in terms of statistical features computed using the Multi-scale Geometric Analysis (MGA) of Non-subsampled Contourlet Transform (NSCT). Noise resilience is one of the main advantages of this feature representation. To improve the retrieval performance and reduce the semantic gap, our system incorporates a Relevance Feedback (RF) mechanism that uses a graph-theoretic approach to rank the images in accordance with the user's feedback. First, a graph of images is constructed with edges reflecting the similarity of pairs of images with respect to the proposed feature representation. Then, images are ranked at each feedback round in terms of the probability that a random walk on this graph reaches an image tagged as relevant by the user before hitting a non-relevant one. Experimental analyses on three different databases show the effectiveness of our algorithm compared to state-of-the-art approaches in particular when the images are corrupted with different types of noise.
M. Chowdhury, Sudeb Das and M. Kundu
In this article, we have proposed a novel Content Based Image Retrieval (CBIR) system, where each image in the database is represented by a compact image signature which is computed using the Nonsubsampled Contourlet Transform (NSCT) and Fuzzy-C-means (FCM) technique. To improve the retrieval accuracy, the proposed system incorporates Least Square Support Vector Machine (LS-SVM) based classifier, Earth Mover's Distance (EMD) and Relevance Feedback Mechanism (RFM). Extensive experiments were carried out to evaluate the effectiveness of the proposed system on SIMPLIcity image database consisting of 1000 images. Experimental results and comparisons show that the proposed CBIR system performs efficiently in image retrieval domain.
Manish Chowdhury, Sudeb Das, and Malay K. Kundu
Springer Berlin Heidelberg
We present a novel Content Based Image Retrieval (CBIR) scheme for natural color images using Multi-scale Geometric Analysis (MGA) of Ripplet Transform (RT) Type-I in the statistical framework based on Generalized Gaussian Density (GGD) model and Kullback- Leibler Distance (KLD). The system is based on modeling the marginal distributions of RT coefficients by GGD framework and computing the similarity between the model parameters using the KLD. Least Square- Support Vector Machine (LS-SVM) classifier is used to classify the im- ages of the database. Extensive experiments were carried out to evaluate the effectiveness of the proposed system on two image databases con- sisting 1000 (Simplicity) and 2788 (Oliva) images, respectively. Exper- imental results and comparisons show that the proposed CBIR system performs efficiently in image retrieval field.
Sudeb Das, Manish Chowdhury, and Malay Kumar Kundu
The Electromagnetics Academy
We propose an automatic and accurate technique for classifying normal and abnormal magnetic resonance (MR) images of human brain. Ripplet transform Type-I (RT), an e-cient multiscale geometric analysis (MGA) tool for digital images, is used to represent the salient features of the brain MR images. The dimensionality of the image representative feature vector is reduced by principal component analysis (PCA). A computationally less expensive support vector machine (SVM), called least square-SVM (LS-SVM) is used to classify the brain MR images. Extensive experiments were carried out to evaluate the performance of the proposed system. Two benchmark MR image datasets and a new larger dataset were used in the experiments, consisting 66, 160 and 255 images, respectively. The generalization capability of the proposed technique is enhanced by 5 £ 5 cross validation procedure. For all the datasets used in the experiments, the proposed system shows high classiflcation accuracies (on an average > 99%). Experimental results and performance comparisons with state-of-the-art techniques, show that the proposed scheme is e-cient in brain MR image classiflcation.
Manish Chowdhury, Sudeb Das, and Malay Kumar Kundu
IEEE
This paper presents a Content Based Medical Image Retrieval (CBMIR) system for diverse collection of radiographic images. Non-Subsampled Contourlet Transform (NSCT) and Fuzzy C-Means (FCM) technique is used to construct the image signature which is used as the image representative feature vector. Least Square-Support Vector Machine (LS-SVM) and Earth Mover's Distance (EMD) is used to classify the images. Preliminary studies on a radiographic image Database (DB) consisting 1550 images of 31 different modalities show promising result.
Malay K. Kundu, Manish Chowdhury, and Minakshi Banerjee
Springer Science and Business Media LLC
We propose an interactive content based image retrieval (CBIR) system using M-band wavelet features with earth mover’s distance (EMD). A fuzzy relevance feedback (FRF) method is proposed to enhance the retrieval mechanism in order to retrieve more images which are semantically close to the query. M × M sub-bands coefficient are used as primitive features, on which, for each pixel, energies are computed over a neighborhood and are taken as features for each pixel to characterize its color and texture properties. Based on the energy property, pixels are clustered using fuzzy C-means algorithm to obtain an image signature. The EMD is used as a distance measure between the signatures for different images of the database. Combining information both from relevant and irrelevant images marked by the user, fuzzy entropy based feature evaluation mechanism is used for automatic computation of revised feature importance and similarity distance at the end of each iteration. The proposed CBIR system performance using M-band wavelets feature are compared to that of Moving Picture Expert Group-7 visual features which have almost become a standard benchmark for both video and image representation and comparison. The proposed FRF technique using EMD is compared with different other similarity measures to test the effectiveness of the system on standard image database.
Manish Chowdhury, Sudeb Das, and Malay Kumar Kundu
Springer Berlin Heidelberg
In this article, a novel content based image retrieval (CBIR) system based on a new Multiscale Geometric Analysis (MGA)-tool, called Ripplet Transform Type-I (RT) is presented. To improve the retrieval result, a fuzzy relevance feedback mechanism (F-RFM) is also implemented. Fuzzy entropy based feature evaluation mechanism is used for automatic computation of revised feature's importance and similarity distance at the end of each iteration. Experimental results on a large image database demonstrate the efficiency and effectiveness of the proposed CBIR system in the image retrieval paradigm
Manish Chowdhury, Sudeb Das, and Malay Kumar Kundu
Universitat Autonoma de Barcelona
Content Based Image Retrieval (CBIR) system is an emerging research area in effective digital data management and retrieval paradigm. In this article, a novel CBIR system based on a new Multiscale Geometric Analysis (MGA)-tool, called Ripplet Transform Type-I (RT) is presented. To improve the retrieval result and to reduce the computational complexity, the proposed scheme utilizes a Neural Network (NN) based classifier for image pre-classification, similarity matching using Manhattan distance measure and relevance feedback mechanism (RFM) using fuzzy entropy based feature evaluation technique. Extensive experiments were carried out to evaluate the effectiveness of the proposed technique. The performance of the proposed CBIR system is evaluated using a 2 £ 5-fold cross validation followed by a statistical analysis. The experimental results suggest that the proposed system based on RT, performs better than many existing CBIR schemes based on other transforms, and the difference is statistically significant.
Sudeb Das, Manish Chowdhury, and Malay Kumar Kundu
The Electromagnetics Academy
The motivation behind fusing multimodality, multi- resolution images is to create a single image with improved interpretability. In this paper, we propose a novel multimodality Medical Image Fusion (MIF) method, based on Ripplet Transform Type-I (RT) for spatially registered, multi-sensor, multi-resolution medical images. RT is a new Multi-scale Geometric Analysis (MGA) tool, capable of resolving two dimensional (2D) singularities and representing image edges more e-ciently. The source medical images are flrst transformed by discrete RT (DRT). Difierent fusion rules are applied to the difierent subbands of the transformed images. Then inverse DRT (IDRT) is applied to the fused coe-cients to get the fused image. The performance of the proposed scheme is evaluated by various quantitative measures like Mutual Information (MI), Spatial Frequency (SF), and Entropy (EN) etc. Visual and quantitative analysis shows, that the proposed technique performs better compared to fusion scheme based on Contourlet Transform (CNT).