1. Ph.D Theses
Permanent URI for this collectionhttps://idr.nitk.ac.in/handle/1/11
Browse
5 results
Search Results
Item Automatic Detection of Malignancy in Low-Magnification Effusion Cytology Images(National Institute of Technology Karnataka, Surathkal, 2024) Aboobacker, Shajahan; Deepu, Vijayasenan; Sumam David S.This thesis focuses on developing an integrated system for automatically detecting malignancy in effusion cytology images. Effusion cytology plays a crucial role in diagnosing various diseases, including cancer, by analyzing the cells present in body fluids. This study aims to develop an integrated system that can handle different resolutions of images and accurately detect malignant cases. Effusion cytology greatly benefits from the automatic detection of malignant cells, providing significant assistance to cytopathologists. However, conventional automation algorithms often rely on high-magnification images for analysis. In contrast, cytopathologists consider multiple magnifications when evaluating cytology images for malignancy. Lower magnification images capture a larger area in a single frame compared to high magnification images. This allows cytopathologists to identify regions of interest (ROI) using textural and morphological characteristics of cell clusters. Once identified, the ROIs are examined at a higher magnification for a closer evaluation at the cell level. Initially, we trained the existing state-of-the-art models with high magnification images for the semantic segmentation and classification of effusion cytology images. We obtained state-of-the-art results for the semantic segmentation task with a mean F-score of 0.82 and classification performance with a sensitivity value of 1, specificity of 0.85, and an area under the curve (AUC) of 0.98. However, using lower magnification images can be beneficial in identifying malignant areas, as it reduces memory requirements and scanning time by focusing only on the ROI at higher magnification. However, detecting malignancy in low-magnification images (4X) is challenging due to the blurring of features such as texture and nuclei. This blurring also makes it difficult to label the images accurately at low magnification levels. Therefore, an alternative method is needed that doesn’t rely on labels for the lowest magnification. We propose two methods for the semantic segmentation of low-magnification images. The first method is based on semi-supervised learning, and the second uses a combination of unsupervised, few-shot and weakly supervised learning. In the semi-supervised approach, we have extended the MixMatch and Fixi Match algorithms from the classification task to semantic segmentation. We used augmentation of the images and reverse augmentation of the predicted label to achieve this. The proposed methods allow using the 4X images without any labels along with the labelled 10X images to train the semantic segmentation model. The average F-score of benign and malignant pixels on the predictions of 4X images using the Extended FixMatch and Extended MixMatch has improved approximately by 9% compared with the predictions of 4X data on the semantic 10X model. The Extended MixMatch reduces the area to be scanned at a higher magnification by approximately 62%. Only 38% of sub-regions of low-magnification images have to be scanned at a higher magnification, thereby saving scanning time. In the context of semi-supervised learning for semantic segmentation of low-magnification images, it is worth noting that while we can reduce the reliance on pixel-wise labels for 4X magnification data, we still require labelled data at a higher magnification level, specifically at 10X. We propose WeakSegNet, a novel approach that combines unsupervised, few-shot and weakly supervised methods for the semantic segmentation of low-magnification effusion cytology images. By leveraging image-level labels and a small number of images with pixel-wise labels, our model achieves accurate and efficient detection of malignancy. Our approach utilizes unlabeled low-magnification images for training, reducing the need for manual annotations. The significant elimination (approximately 47%) achieved by our model in higher magnification scanning demonstrates its potential for time and resource savings. Overall, our approaches offer an effective solution for automating malignancy detection in low-magnification images, improving efficiency in cytology analysis.Item Soil Fertility Classification Using Machine Learning-Based Approach(National Institute of Technology Karnataka, Surathkal, 2024) M. Sujatha; D.Jaidhar C.Agriculture is the main source of economy and survival in many countries. To ensure sustainable agricultural development, it is crucial to promptly acquire soil fertility and apply accurate fertilizers. However, traditional laboratory methods for analyzing soil samples make it challenging to estimate soil fertility. Therefore, this research aims to develop a reliable Machine Learning (ML)-based classifier that can classify soil fertility as LOW, or MEDIUM, or HIGH. Additionally, prescribes fertilizers based on the classification results. Soil fertility classification approach based on laboratory chemical parameters such as Electrical Conductivity (EC), Organic Carbon (OC), potential of hydrogen (pH), boron (B), copper (Cu), iron (Fe), manganese (Mn), phosphorus (P), potassium (K), sulphur (S), and zinc (Zn) have been proposed using ML approaches. The classifiers used in this study included Random Forest (RF), bagging, Boosted Regression Tree (BRT), J48 Decision Tree (J48), Logistic Regression (LR), Naive Bayes (NB), and Support Vector Machine (SVM). The experiments were conducted with a split dataset (75% of data for training and 25% for testing) and 10-fold cross-validation. The tree-based classifier RF, outperformed the other classifiers by producing an accuracy of 99.99% with 10-fold cross-validation test and a split dataset. To avoid the need for laboratory analysis and obtain soil parameters specific to the site, this research relied on Sentinel-2 spectral data to determine EC, pH, OC, and N. The generated dataset was labeled using various clustering methods such as canopy, density-based, expectation-maximization, farthest-first, fuzzy C-means, and k-means and then compared with manual labeling. Among these, the canopy clustering approach achieved the highest accuracy of 75.99% on labeling dataset. Therefore, the proposed method for labeling the dataset uses the canopy-centered fuzzy C-means clustering. It was found that the proposed canopy-centered fuzzy-C-means clustering method achieved the highest accuracy of 78.42% in labeling the dataset. Furthermore, the performance of several ML-based classifiers, such as NB, SVM, J48, and RF were compared using datasets labeled with different clustering approaches. The RF classifier achieved the highest classification accuracy of 99.69% using the proposed approach and on 10-fold cross-validation. To determine the best fertilizer for a given soil, a new fertilizer prescription approach was proposed. It uses an ensemble filter-based feature selection to classify soil fertility and prescribe the appropriate fertilizer. It was tested on two datasets from regions with varying climate conditions. Various tree-based classifiers, such as classification and regression tree, extra tree, reduced error pruning tree, RF, NB, and SVM, were compared using the first dataset with relevant soil parameters. The results showed that the RF classifier with relevant soil parameters was the most accurate, achieving a 99.96% i accuracy with dataset-1 and a 99.90% accuracy with dataset-2. A soil fertility classifier and fertilizer prescription approach was proposed by utilizing 2D Convolutional Neural Networks (CNNs). The experiments were conducted on a split dataset with varying kernel sizes of 3×3 to 7×7 and input grid sizes from 11×11 to 13×13. The classifier showed an impressive accuracy of 97.24% and kappa statistics of 0.0938 with an input grid size of 11×11 and a kernel size of 3×3. To further improve the accuracy, the training data was oversampled using the Synthetic Minority Oversampling Technique (SMOTE). The proposed approach using oversampling achieved an accuracy of 97.52% and kappa statistics of 0.1397, with an input grid size of 12×12 and a kernel size of 3×3. A 1D-CNN-based soil fertility classification approach was developed to simplify the 2D CNN-based classifier used for soil fertility classification. To improve the performance of the model, the dataset was normalized using Min-Max normalization, and training data was oversampled using SMOTE. The proposed approach was compared with the soil fertility classifiers based on Extreme Learning Machine (ELM) and Multi- Layer Perceptron (MLP). The proposed approach, with normalization and SMOTE, achieved an accuracy of 97.90% and kappa statistics of 0.2358. A new method to classify soil fertility and prescribe fertilizers using symbolic deterministic finite automata, to overcome the limitations of traditional ML-based classifiers, which require large, unbiased datasets and are prone to errors. The proposed method was compared using ML-based classifiers using data from Sentinel-2 satellite imagery and laboratory-measured soil health data of Belgaum district. The data consisted of two sets: one with four soil parameters (Soil-health-1 dataset) and the other with twelve soil parameters (Soil-health-2 dataset). The results showed that the new approach was able to classify soil fertility with 100% accuracy using the Sentinel-2 and Soil-health-1 datasets, and with 98.37% accuracy using the Soil-health-2 dataset. Satellite revisits to a specific site location are infrequent, hence, soil sensors are used to collect real-time values of EC, pH, N, P, and K in this study. The collected real-time data is tested using trained and saved ML-based classifiers, such as Classification and Regression Tree (CART), J48, RF, Reduced Error Pruning (REP), NB and SVM which were trained using the Soil-health dataset of Belgaum district. For the real-time test data RF and REP classifiers achieved highest test accuracy of 100%.Item Early Diagnosis of Osteoporosis using Metacarpal Radiogrammetry and Texture Analysis(National Institute of Technology Karnataka, Surathkal, 2019) Areeckal, Anu Shaju.; S, Sumam DavidOsteoporosis is a disease characterized by reduction in bone mass and micro-structure, leading to increased risk of fragility fractures. The gold standard technique used for diagnosis of osteoporosis is determination of Bone Mineral Density (BMD) using Dual Energy X-ray Absorptiometry (DXA). DXA is accurate and precise, however it has a high cost of scan and low availability in developing countries. The aim of the research work is to develop a low cost prescreening tool for early diagnosis of osteoporosis using cortical radiogrammetry and trabecular texture analysis of hand and wrist radiographs. An automatic method for segmentation of the third metacarpal bone shaft from hand and wrist radiographs is proposed using automatically detected anatomical landmarks, intensity profiles and marker-controlled watershed segmentation. From the outer and inner bone edges of the segmented third metacarpal bone shaft, cortical radiogrammetric features are extracted. The proposed method is validated on sample data of two ethnic groups: 138 Indian subjects and 65 Swiss subjects. The proposed segmentation method accurately detected the third metacarpal bone in 89% of Indian sample data and 78% of Swiss sample data. The proposed method shows better performance as compared with the state-of-the-art segmentation method, Active Appearance Model (AAM). A segmentation approach is proposed to automatically extract the distal radius for trabecular texture analysis. The proposed method extracted the distal radius region-ofinterest in 93.5% of Indian sample data and 83% of Swiss sample data. Texture analysis of the trabecular bone in distal radius is done. The extracted features are analyzed using independent sample t-test and Pearson correlation analysis. The cortical radiogrammetric features show high discrimination ability in the healthy and low bone mass groups of both Indian and Swiss sample data. The cortical and texture features are divided into different feature sets. Classifiers are trained on cortical features and statistical and structural texture features for Inviidian sample data and a linear regression model is estimated. Artificial Neural Network (ANN) classifier trained using holdout validation achieves test accuracy of 90.0%. kNearest Neighbor (KNN) using 10-fold cross validation achieves an accuracy of 81.7%. The linear regression model developed with the cortical and texture features achieves a significant correlation of 0.671 with DXA-BMD. Classifiers are also trained separately for Indian and Swiss sample population. ANN classifiers trained with significant cortical and statistical and structural texture features show test accuracy of 92.9% with Indian data and 90.9% with Swiss data. Weighted KNN using the same feature set shows test accuracy of 96.2% using holdout validation. A novel method to measure the cortical volume of the metacarpal bone shaft at a low cost using three dimensional reconstruction from hand X-ray images in three views (Postero-Anterior, 450 and 1350 oblique views) is proposed. The Computed Tomography scan of one subject is used to create a template model, from which subject-specific models of other subjects are reconstructed. The 3D reconstruction of the bone is done iteratively by registration of projection and X-ray contours using Iterative Closest Point and Self-Organizing Map, and deformation of the template model using Laplacian surface deformation. The outer and inner bone walls of the metacarpal are modeled separately and the third metacarpal bone shaft is extracted from which cortical volume is measured. The projections of the 3D reconstructed models are compared with manually segmented X-ray images and the mean percentage error in Combined Cortical Thickness (CCT) is 11.18%. In summary, a low cost prescreening tool for early diagnosis of osteoporosis using cortical radiogrammetry and texture analysis is proposed and validated using sample data of Indian and Swiss population. A low cost method to measure cortical volume of third metacarpal bone shaft using multi-view hand radiographs is also proposed. This work is done in collaboration with Kasturba Medical College Hospital, Mangalore, India and University Hospital of Geneva, Switzerland.Item Development of CAD System for Detection, Classification, Retrieval and 3d Reconstruction of Brain and Liver Tumors on MRI and CT Images(National Institute of Technology Karnataka, Surathkal, 2014) Arakeri, Megha P.; Reddy, G. Ram MohanaBrain and liver tumors are the life-threatening diseases due to low survival rate. Hence, accurate diagnosis of brain and liver tumors is necessary to provide effective treatment. Medical imaging techniques like Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) help in acquiring images of the tumor. The visual analysis of these medical images by the radiologist is time consuming, subjective and inaccurate. The needle biopsy of the tumor provides accurate diagnosis but it is an invasive technique and generally not recommended. In order to overcome these drawbacks, there is a need to develop Computer-Aided Diagnosis (CAD) system for assisting the radiologist in fast and accurate diagnosis of tumors. Therefore, this thesis proposes an effective and efficient CAD system for tumor detection, classification, Content-Based Image Retrieval (CBIR), and 3D reconstruction to provide complete assistance to the radiologist in the diagnosis of brain and liver tumors. In the first methodology, this thesis aims at tumor detection, by proposing automatic, effective and efficient segmentation methods for brain and liver tumors on medical images. The brain tumor is detected using the proposed segmentation technique based on Modified Fuzzy C-Means (MFCM) clustering algorithm. The liver tumor is detected using the proposed segmentation technique based on the automatic region growing algorithm. In the second methodology, this thesis targets at identification of the type of brain/liver tumor as benign or malignant, by proposing an effective and efficient tumor classification scheme. Precisely, the proposed scheme represents the tumor characteristics using its significant features, selects most discriminating features using a two-level feature selection technique consisting of Information Gain (IG) based feature ranking and Independent Component Analysis (ICA) based feature section methods. Then, the tumor is classified using an ensemble classifier consisting of Support Vector Machine (SVM), Artificial Neural Network (ANN) and k-Nearest Neighbor (k-NN) classifiers. In the third methodology, this thesis proposes two CBIR methods based on image rotation correction and rotation invariant features to assist the radiologist in brain/liver tumor diagnosis based on past resolved cases. In order to provide fast retrieval of tumor images from the database, the tumor features in the database are indexed using the proposed indexing technique called as Cluster with IG-ICA and KD-tree (CIKD). The features in the database are partitioned into different groupsusing modified k-means clustering which identifies the number of clusters and initial cluster centers automatically. In the fourth methodology, this thesis aims to build the 3D model of the brain/liver tumor, by proposing an effective and efficient 3D reconstruction scheme. Precisely, it proposes an enhanced shape-based interpolation algorithm to estimate missing slices in a given set of brain/liver tumor slices. Further, the 3D mesh simplification algorithm is proposed to reduce the number of triangles in the reconstructed mesh and accelerate the rendering phase. The tumor volume is also computed to assist the radiologist in estimating the stage of cancer. Experiments are carried out on a dataset consisting of MRI images of the brain tumor and CT images of the liver tumor. Experimental results demonstrate that the proposed CAD system is automatic, effective and efficient in the diagnosis of brain and liver tumors.Item Effective Multimedia Document Representations for Knowledge Discovery(National Institute of Technology Karnataka, Surathkal, 2017) K, Pushpalatha; V. S, AnanthanarayanaIn recent years, the rapid advances in multimedia technology have led to grow the multimedia documents explosively. In order to utilize the multimodal information of multimedia documents, sophisticated knowledge discovery systems are required. The knowledge discovery systems require efficient multimedia mining methods to extract the meaningful and useful information from the huge volume of multimedia documents. The success of multimedia mining relies on the representation of multimedia documents and its multimodal contents. The appropriate representation of multimedia documents discovers the useful patterns that can be used to assist the multimedia mining methods in discovering the useful knowledge. The multimodal nature of multimedia objects is the challenging problem for the multimedia document representation, as the features of multimodal objects are in different space with different characteristics and dimensionalities. Representation of multimodal multimedia objects in a unified feature space helps the multimedia document representation and multimedia mining methods. The research work in this thesis proposes the multimedia data representation methods, multimedia document representations, and multimedia mining methods for the effective knowledge discovery in multimedia documents. In the first methodology, this thesis aims at the representation of multimodal multimedia objects in a unified feature space. We propose two multimedia data representation methods, Multimedia To Signal Conversion (MSC) and Multimedia to Image Conversion (MIC) to represent the multimedia objects in a unified domain. The MSC represents the multimedia objects in frequency domain by converting the multimedia objects as signal objects. The MIC converts the multimedia objects as image objects to represent them in spatial domain. The multimedia objects in unified domain are represented in the unified feature space using the features with similar dimensions and characteristics. Hence, both the multimedia data representation methods convert themultimodal multimedia documents as unified multimedia documents. The unified multimedia documents ease the representation of multimedia documents and improve the efficiency of multimedia mining methods. The proposed multimedia data representation methods are effectively used for knowledge extraction from multimedia documents. In the second methodology, this thesis presents the two multimedia document representations, Multimedia Suffix Tree Document (MSTD) and Multimedia Feature Pattern Tree (MFPT) to represent the unified multimedia documents. The MSTD represents the unified multimedia documents based on shared similar multimedia objects among the documents. The similarity between the multimedia objects depends on the similarity of the features. The MFPT represents the documents based on shared similar feature patterns of the multimedia objects. Both the representations are compact and provide the complete information of the documents. They function as the platform for the multimedia knowledge extraction methods. In the third methodology, this thesis explores the multimedia mining methods based on the MSTD and MFPT representations. The MSTD and MFPT based classification algorithms effectively classifies the multimedia documents. The multimedia documents are partitioned into clusters of same multimedia concepts using the MSTD and MFPT based clustering algorithms. The MSTD representation extracts the frequent multimedia patterns to generate the multimedia class association rules for classifying the multimedia documents. The MFPT representation extracts the sequential multimedia feature patterns to derive the multimedia class sequential rules that support the classification of multimedia documents based on the object characteristics. The efficacy of the proposed methods is evaluated by conducting the experiments with four datasets of multimodal multimedia documents. Experimental results demonstrate that the proposed multimedia data representation methods benefit the multimedia document representation and multimedia mining methods by representing the multimodal multimedia objectsin a unified feature space. The proposed multimedia document representations are effectively used to enhance the performance of multimedia mining methods in discovering the knowledge from multimedia documents.