RESEARCH AREA & SUMMARY

 

  • Biostatistics & Medical Informatics
  • Computational Pathology & Neuroinformatics
  • Statistical Machine Learning & Computer Aided Diagnosis (CAD)
  • Computer Vision & Pattern Recognition for Medical Imaging Informatics

BRIEF OUTLINE OF MY RESEARCH WORK

Disease Screening Method & System Development

Globally, recent advancements of biomedical image analysis, statistical modeling and pattern recognition techniques have been strengthening modern quantitative macro-microscopy in overcoming the subjectivity in the visual assessment of morphological, textural and intensity based characteristics at tissue as well as cellular levels. The reason behind disease screening tool development is to screen the cases from population specially onsite (a) when pathologist and doctor are not available; (b) cases should be quickly screened for reporting to referral hospital; (c) finally quantitative estimation of diagnostic markers for more accurate clinical decision making.
Towards this direction, the interdisciplinary research works viz.,  development of computer-assisted diagnostic methodology for screening oral submucous fibrosis (OSF), malaria, leukemia and breast carcinoma and polycystic ovary syndrome using medical image processing, statistical  analysis and pattern recognition techniques have been undertaken in Biostatistics & Medical Informatics (BMI) Laboratory at School of Medical Science & Technology, IIT Kharagpur.

Oral cancer screening using histopathological images

Oral Submucous Fibrosis (OSF) is a precancerous condition leading to oral malignancy of human being because of late diagnosis. For OSF screening, microscopic images of stained histopathological samples from oral biopsies are preprocessed using median filter and histogram based contrast enhancement. This study investigates the morphological, textural and intensity-based variations in epithelium, sub-epithelium and basement membrane from normal to OSF. Watershed segmentation has been extended to segment the epithelial layer based on the combination of texture and color gradients, which shows better result than Otsu’s thresholding. Further, the layer is quantified for characterizing thickness and texture. Moreover, the texture features (wavelet, Gabor-wavelet, local binary pattern, fractal dimension) have been combined for grading histopathological tissues into normal, OSF with and without dysplasia. His group designed a hybrid segmentation consisting of fuzzy divergence, color de-convolution and gradient vector flow for extracting basal cell nuclei. Morphology (area, perimeter, eccentricity, convexity, form factor etc) and texture of these nuclei are analyzed at 100x magnification. Another significant microscopic characteristic of OSF is the deposition of collagen in the sub-epithelial connective tissue (SECT) leading to epithelial atrophy. He proposed a methodology for segmenting collagen fibers using neural network and its textural feature extraction using fractal dimension. Being responsible for oral malignancy, SECT cell population are filtered out using color de-convolution and differentiated into inflammatory and fibroblast cells using Zernike moments and Fourier descriptors. Statistical analysis showed that fibroblast cells have increasing trend to cause fibrosis for OSF. 95 features are found to be statistically significant out of total 103. Thereafter, supervised (SVM, Bayesian, Neural network) and unsupervised (K-means, FCM, GMM) classifiers are trained based on 45 OSF cases with more than 85% screening accuracy.

Blood pathological image analysis and visualization

In order to develop a screening method for malaria, anaemia and leukemia, the microscopic images of peripheral blood smears are grabbed at 10x and 100x magnifications targeting morphology and texture of erythrocytes and leukocytes. Fuzzy and intuitionistic fuzzy divergence measures have been proposed using Renyi and Yager’s entropy functions and compared with Shanon’s entropy for segmentation of cells and nuclei. Leukocytes are characterized using morphological and textural features where 8 features including nucleo-cytoplasmic ratio are found to be statistically significant. Five types of leukocytes are automatically recognized using Bayesian classifier with 83.2% accuracy. For malaria, textural features viz., average intensity, skewness, uniformity, entropy, fractals are found to be significant in delineating Plasmodium vivax infected region of erythrocytes.

Breast cancer screening using FNAC & X-ray Mammogram images

Method and system development for analyzing breast carcinoma using microscopic images of fine needle aspirates has been attempted. The method included extracting a G-plane image and de-noising speckle and salt-paper noise using median filter. Furthermore, the method generated a binary image from G-plane image using Otsu’s thresholding. It filtered the binary image to yield nuclear map from which nuclear contour is extracted. The microscopic features of nuclei viz., radius, perimeter, area, compactness, smoothness, texture are extracted from the contour and statistically analyzed for classifying breast lesion into either malignant or non-malignant.

Mass screening of PCOS using clinical and imaging markers

Recently, I have been involved in a burning research problem on characterization of polycystic ovary syndrome (PCOS) in women of reproductive age (20-35 yrs) using clinical, hormonal and ultrasound imaging parameters in collaboration with Ghosh Dastidar Institute for Fertility Research, Kolkata. The clinical and hormonal information are being statistically analyzed to identify the pattern of these parameters between normal and PCOS women. As Ultrasound and color doppler  images involve lot of ambiguities and speckle noises, it is very important to preprocess for not only better visualization but also precise quantification of target features in respect to ovary, follicle and endometrium.

Diabetic retinopathy screening using fundus images

There are varieties of retinal disorders like retinopathy, maculopathy, glaucoma etc that affect our population vary much as a consequence of change of lipid and hormonal profile of the human body. Advanced imaging technology viz., OCT integrated with Fundus camera strengthens the registration and characterization of retinal images for precise understanding of the disease pathogenesis prior to therapeutic intervention.

Coronary Artery Disease Screening

Coronary artery diseases (CAD) have become very common, especially to Indian population, and leading cause of death.  The incidence rate is rising at alarming rate because of several risk factors viz., hypertension, diabetes, higher cholesterol level, smoking, and also junk foods etc. In such situation, preventive measures should be attempted by screening frequently the subjects in order to reduce the mortality rate. Under these circumstances, automated cardiac screening tool development is now a major challenge. To develop such screening tool, the following objectives are to be set as follows:

(a) Design and development of clinical questionnaire and database for CAD in
      consultation with cardiologists.
(b)  Development of databases on ECG and PCG.
(c)  Pre-processing of ECG and PCG both signals by noise reduction.
(d)  Selection of target feature from signals by cardiologist.
(e)  Feature extraction from ECG and PCG as per target features.
(f)   Statistical modelling and analysis.
(g)  Extraction of new parameters by correlating simultaneously recorded ECG & PCG
       signals.
(h)  Development of robust supervised and unsupervised classification techniques.

Health Informatics with Statistical Analytics

Development of Portable, Low-Cost Imaging for Monitoring and Disease Diagnosis

Microscopic imaging is a vital and omnipresent healthcare tool in modern hospitals and clinics for initial disease screening as well as for in-depth analysis of patient blood samples. Especially in remote places of our country, experts like pathologist, dermatologist etc is not available for instant report generation. In such situation, a patient has to wait for some days to get report in hand before going to a doctor. In fact, remote places lack access to clinical-quality microscopes necessary for even the most basic evaluations. This scarcity of equipment is exacerbated by the lack of qualified medical personnel, especially in rural areas, to provide diagnosis and treatment based on the microscopic data.
In view of this, the objective of this work is to develop a modular, high-magnification microscope attachment for cell phones. Due to its portability, affordability and functionality, the cell phone based microscope will enable health workers in remote areas to take high-resolution images of a patient's blood cells and skin patches using the mobile phone's camera, and then transmit the photos to experts at medical centers. This device can reduce both the cost and time of performing critical disease diagnoses, as well as provide early warning of outbreaks in poverty-striking regions. From automated diagnostics point of view, it is expected to image malaria parasites and tuberculosis bacteria etc through development and deployment of robust image analysis and pattern classification algorithms.

Distributed database and Statistical Analytics

All diagnostic databases like pathology, radiology, biochemistry, hematology databases will be connected though distributed database protocol for central health information system. Simultaneously, a statistical analyzer based on important statistical methods needs to be developed for exploratory analysis, trend analysis of diseases, malnutrition and other health hazards on the basis of collected data.  The inference for preventive healthcare will be done based on the statistical techniques.

  Copyright 2011. All Rights Reserved