Mita Nasipuri
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Mita Nasipuri has a total of 135 co-authors. Here are the top co-authors by number of shared publications:
- Debotosh Bhattacharjee
Department of Computer Science and Engineering, Jadavpur University, Kolkata, India
47 shared publications - Subhadip Basu
Jadavpur University
28 shared publications - Ram Sarkar
Computer Science and Engineering Department, Jadavpur University, Kolkata, India
18 shared publications - Dipak Kumar Basu
Jadavpur University
15 shared publications - Nibaran Das
Department of Computer Science and Engineering, Jadavpur University, Kolkata, India
13 shared publications - Anup Kumar Halder
Department of Computer Science and Engineering, Indian Institute of Information Technology, Raichur, Karnataka, India
11 shared publications
Latest Publications
- 12 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-14
2pCePd-Net: Two-Path Cross-Context Encoder With Probability Map-Based Bandpass Decoder for Retinal Vessel Segmentation
Accurate automatic retinal blood vessel segmentation in fundus images plays an important role in the early diagnosis of any ocular disease detection system. However, most of the past literature has yet to attain a superior result primarily due to a lack of sufficient annotated data and the complexity of the vessel structure under challenging background conditions. In this work, we propose the design of a coherence measure-guided data augmentation model, named lambda-coherence measure-guided Cartesian-square ( $\lambda \text {CMgC}^{2}$ ), for enriching the existing datasets with synthetic and structurally coherent fundus images thus alleviating the issue of data scarcity. Subsequently, we propose a novel end-to-end convolutional network, called two-path cross-context encoder with probability map-based bandpass decoder (2pCePd-Net) for the segmentation of blood vessels endowed with a novel $\text {2pCd}^{+}$ encoder block with CERg skip connection and a novel $\hat {p}\text {BPf}$ -enabled decoder block. The proposed work has been evaluated using four standard datasets, namely, DRIVE, STARE, CHASEDB1, and HRF, and has obtained a benchmark accuracy (Ac) of 97.6%, 98.1%, 98.2%, and 97.7%, respectively. Statistically, our proposed model has further achieved benchmark results across sensitivity (Se), specificity (Sp), $F1$ , and AUC measures of evaluation as well.
- 12 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-14
2pCePd-Net: Two-Path Cross-Context Encoder With Probability Map-Based Bandpass Decoder for Retinal Vessel Segmentation
Accurate automatic retinal blood vessel segmentation in fundus images plays an important role in the early diagnosis of any ocular disease detection system. However, most of the past literature has yet to attain a superior result primarily due to a lack of sufficient annotated data and the complexity of the vessel structure under challenging background conditions. In this work, we propose the design of a coherence measure-guided data augmentation model, named lambda-coherence measure-guided Cartesian-square ( $\lambda \text {CMgC}^{2}$ ), for enriching the existing datasets with synthetic and structurally coherent fundus images thus alleviating the issue of data scarcity. Subsequently, we propose a novel end-to-end convolutional network, called two-path cross-context encoder with probability map-based bandpass decoder (2pCePd-Net) for the segmentation of blood vessels endowed with a novel $\text {2pCd}^{+}$ encoder block with CERg skip connection and a novel $\hat {p}\text {BPf}$ -enabled decoder block. The proposed work has been evaluated using four standard datasets, namely, DRIVE, STARE, CHASEDB1, and HRF, and has obtained a benchmark accuracy (Ac) of 97.6%, 98.1%, 98.2%, and 97.7%, respectively. Statistically, our proposed model has further achieved benchmark results across sensitivity (Se), specificity (Sp), $F1$ , and AUC measures of evaluation as well.
- 12 February 2025
- Published by Springer Nature
- p. 257-269
Regularizing CNNs Using Confusion Penalty Based Label Smoothing for Histopathology Images
- 12 February 2025
- Published by Springer Nature
- p. 257-269
Regularizing CNNs Using Confusion Penalty Based Label Smoothing for Histopathology Images
- 12 February 2025
- Published by Springer Nature
- p. 100-112
COVID-CT-H-UNet: A Novel COVID-19 CT Segmentation Network Based on Attention Mechanism and Bi-Category Hybrid Loss
- 12 February 2025
- Published by Springer Nature
- p. 100-112
COVID-CT-H-UNet: A Novel COVID-19 CT Segmentation Network Based on Attention Mechanism and Bi-Category Hybrid Loss
- 7 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Biology and Bioinformatics
- Vol. 22 (3) , 997-1008
PCPredG: Protein Complex Prediction Using Graphlet Features
Proteins interact with other proteins and bio-molecules to form a complex and execute key biological functions in a living organism, and respond to several environmental signals. Designing efficient predictive models for protein complexes is a challenging task with limited coverage in the contemporary literature. With this motivation, we have developed a novel method, PCPredG, for 3-node protein complex prediction from PPI networks using 5-node graphlet features. CORUM protein complex repository has been used to curate positive and negative data samples with the help of MCODE and MCL clustering algorithms. During experiments, Random Forest(RF) and SVM classifiers are trained with 1000 positive 3-node complexes in 10- fold cross-validation setup and with 1:1 to 1:10 positive-negative proportions. In parallel, we have implemented the state-of-the-art GCN with polarised message-passing, GAT and an ensemble of GCN and GAT in both balanced and imbalanced setups. We also introduced a 10- fold quality consensus on the hold-out set across all the experiments. We have achieved the best performances with the RF classifier in both balanced and imbalanced experiments. The dataset and the developed PCPredG model are available at https://github.com/CMATERJU-BIOINFO/PCPredG for academic use only.
- 7 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Biology and Bioinformatics
- Vol. 22 (3) , 997-1008
PCPredG: Protein Complex Prediction Using Graphlet Features
Proteins interact with other proteins and bio-molecules to form a complex and execute key biological functions in a living organism, and respond to several environmental signals. Designing efficient predictive models for protein complexes is a challenging task with limited coverage in the contemporary literature. With this motivation, we have developed a novel method, PCPredG, for 3-node protein complex prediction from PPI networks using 5-node graphlet features. CORUM protein complex repository has been used to curate positive and negative data samples with the help of MCODE and MCL clustering algorithms. During experiments, Random Forest(RF) and SVM classifiers are trained with 1000 positive 3-node complexes in 10- fold cross-validation setup and with 1:1 to 1:10 positive-negative proportions. In parallel, we have implemented the state-of-the-art GCN with polarised message-passing, GAT and an ensemble of GCN and GAT in both balanced and imbalanced setups. We also introduced a 10- fold quality consensus on the hold-out set across all the experiments. We have achieved the best performances with the RF classifier in both balanced and imbalanced experiments. The dataset and the developed PCPredG model are available at https://github.com/CMATERJU-BIOINFO/PCPredG for academic use only.
- 1 February 2025
- Published by World Scientific Pub Co Pte Ltd in International Journal on Artificial Intelligence Tools
- Vol. 34 (01)
GA-RISE: Posthoc Model Agnostic Explanations of Black-Box Classifiers Using Genetic Algorithm-Based Optimized Masks — A Case Study on Chest X-ray Images
The intricate and recursive nature of deep learning models often obscures their inner workings, limiting their applications in high-risk domains like medical image analysis. The development of Explainable AI (XAI) methods has addressed this limitation by providing human-readable insights into these black-box models, which are widely used in image-based tasks such as classification, segmentation, and localization. One popular and effective XAI technique is Randomized Input Sampling for Explanation (RISE), which uses a large number of random masks to visualize model behavior. However, RISE’s requirement for numerous masks leads to high space complexity. In this paper, we propose a modified RISE technique that optimizes the use of a relatively small number of masks through a Genetic Algorithm to enhance comprehensibility. Experiments with state-of-the-art classifier models on a publicly available Chest X-ray dataset demonstrate that our approach outperforms other commonly used post hoc model-agnostic and non-model-agnostic techniques in terms of explanation quality and faithfulness metrics.
- 1 February 2025
- Published by World Scientific Pub Co Pte Ltd in International Journal on Artificial Intelligence Tools
- Vol. 34 (01)
GA-RISE: Posthoc Model Agnostic Explanations of Black-Box Classifiers Using Genetic Algorithm-Based Optimized Masks — A Case Study on Chest X-ray Images
The intricate and recursive nature of deep learning models often obscures their inner workings, limiting their applications in high-risk domains like medical image analysis. The development of Explainable AI (XAI) methods has addressed this limitation by providing human-readable insights into these black-box models, which are widely used in image-based tasks such as classification, segmentation, and localization. One popular and effective XAI technique is Randomized Input Sampling for Explanation (RISE), which uses a large number of random masks to visualize model behavior. However, RISE’s requirement for numerous masks leads to high space complexity. In this paper, we propose a modified RISE technique that optimizes the use of a relatively small number of masks through a Genetic Algorithm to enhance comprehensibility. Experiments with state-of-the-art classifier models on a publicly available Chest X-ray dataset demonstrate that our approach outperforms other commonly used post hoc model-agnostic and non-model-agnostic techniques in terms of explanation quality and faithfulness metrics.
- Open Access
- 19 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 1249-1284
Devanagari Character Recognition: A Comprehensive Literature Review
The Devanagari script originated from the ancient Brahmi script and is a widely used Indic script for writing different languages, like Sanskrit, Hindi, Marathi, Nepali, and Konkani. Recognizing handwritten Devanagari characters poses significant challenges due to their complexity and handwriting variability. This literature review examines the evolution of handwritten Devanagari character recognition (HDCR), exploring early template matching and feature extraction methods that struggled with the script’s intricacy. Advances introduced structural and statistical techniques, improving accuracy by analyzing geometric properties and patterns. The advent of machine learning, particularly deep learning, revolutionized HDCR with convolutional neural networks (CNNs) and recurrent neural networks (RNNs), significantly enhancing performance. Hybrid approaches that combine multiple techniques have shown promising results, balancing accuracy and computational complexity. Challenges remain, including handwriting variability, noise, and the need for real-time performance. The lack of large, diverse datasets for training and evaluation is a significant hurdle. This review highlights efforts to create annotated datasets and benchmarks, providing a comprehensive overview of HDCR methodologies, strengths, limitations, and future research directions. These insights aim to advance HDCR, contributing to more accurate and efficient recognition systems and enhancing digital text processing for linguistic, educational, and archival purposes.
- 19 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 1249-1284
Open AccessDevanagari Character Recognition: A Comprehensive Literature Review
The Devanagari script originated from the ancient Brahmi script and is a widely used Indic script for writing different languages, like Sanskrit, Hindi, Marathi, Nepali, and Konkani. Recognizing handwritten Devanagari characters poses significant challenges due to their complexity and handwriting variability. This literature review examines the evolution of handwritten Devanagari character recognition (HDCR), exploring early template matching and feature extraction methods that struggled with the script’s intricacy. Advances introduced structural and statistical techniques, improving accuracy by analyzing geometric properties and patterns. The advent of machine learning, particularly deep learning, revolutionized HDCR with convolutional neural networks (CNNs) and recurrent neural networks (RNNs), significantly enhancing performance. Hybrid approaches that combine multiple techniques have shown promising results, balancing accuracy and computational complexity. Challenges remain, including handwriting variability, noise, and the need for real-time performance. The lack of large, diverse datasets for training and evaluation is a significant hurdle. This review highlights efforts to create annotated datasets and benchmarks, providing a comprehensive overview of HDCR methodologies, strengths, limitations, and future research directions. These insights aim to advance HDCR, contributing to more accurate and efficient recognition systems and enhancing digital text processing for linguistic, educational, and archival purposes.
- Open Access
Fusion Based Hand Geometry Recognition Using Dempster-Shafer Theory
This paper presents a new technique for person recognition based on the fusion of hand geometric features of both the hands without any pose restrictions. All the features are extracted from normalized left and right hand images. Fusion is applied at feature level and also at decision level. Two probability based algorithms are proposed for classification. The first algorithm computes the maximum probability for nearest three neighbors. The second algorithm determines the maximum probability of the number of matched features with respect to a thresholding on distances. Based on these two highest probabilities initial decisions are made. The final decision is considered according to the highest probability as calculated by the Dempster-Shafer theory of evidence. Depending on the various combinations of the initial decisions, three schemes are experimented with 201 subjects for identification and verification. The correct identification rate found to be 99.5%, and the False Acceptance Rate (FAR) of 0.625% has been found during verification.
Open AccessFusion Based Hand Geometry Recognition Using Dempster-Shafer Theory
This paper presents a new technique for person recognition based on the fusion of hand geometric features of both the hands without any pose restrictions. All the features are extracted from normalized left and right hand images. Fusion is applied at feature level and also at decision level. Two probability based algorithms are proposed for classification. The first algorithm computes the maximum probability for nearest three neighbors. The second algorithm determines the maximum probability of the number of matched features with respect to a thresholding on distances. Based on these two highest probabilities initial decisions are made. The final decision is considered according to the highest probability as calculated by the Dempster-Shafer theory of evidence. Depending on the various combinations of the initial decisions, three schemes are experimented with 201 subjects for identification and verification. The correct identification rate found to be 99.5%, and the False Acceptance Rate (FAR) of 0.625% has been found during verification.
- 5 July 2024
- Published by Springer Nature
- p. 457-469
WSports-50: An Image Dataset for Women’s Sport Action Classification
- 5 July 2024
- Published by Springer Nature
- p. 457-469
WSports-50: An Image Dataset for Women’s Sport Action Classification
- Open Access
Regularizing CNNs using Confusion Penalty Based Label Smoothing for Histopathology Images
Deep Learning, particularly Convolutional Neural Networks (CNN), has been successful in computer vision tasks and medical image analysis. However, modern CNNs can be overconfident, making them difficult to deploy in real-world scenarios. Researchers propose regularizing techniques, such as Label Smoothing (LS), which introduces soft labels for training data, making the classifier more regularized. LS captures disagreements or lack of confidence in the training phase, making the classifier more regularized. Although LS is quite simple and effective, traditional LS techniques utilize a weighted average between target distribution and a uniform distribution across the classes, which limits the objective of LS as well as the performance. This paper introduces a novel LS technique based on the confusion penalty, which treats model confusion for each class with more importance than others. We have performed extensive experiments with well-known CNN architectures with this technique on publicly available Colorectal Histology datasets and got satisfactory results. Also, we have compared our findings with the State-of-the-art and shown our method's efficacy with Reliability diagrams and t-distributed Stochastic Neighbor Embedding (t-SNE) plots of feature space.
Open AccessRegularizing CNNs using Confusion Penalty Based Label Smoothing for Histopathology Images
Deep Learning, particularly Convolutional Neural Networks (CNN), has been successful in computer vision tasks and medical image analysis. However, modern CNNs can be overconfident, making them difficult to deploy in real-world scenarios. Researchers propose regularizing techniques, such as Label Smoothing (LS), which introduces soft labels for training data, making the classifier more regularized. LS captures disagreements or lack of confidence in the training phase, making the classifier more regularized. Although LS is quite simple and effective, traditional LS techniques utilize a weighted average between target distribution and a uniform distribution across the classes, which limits the objective of LS as well as the performance. This paper introduces a novel LS technique based on the confusion penalty, which treats model confusion for each class with more importance than others. We have performed extensive experiments with well-known CNN architectures with this technique on publicly available Colorectal Histology datasets and got satisfactory results. Also, we have compared our findings with the State-of-the-art and shown our method's efficacy with Reliability diagrams and t-distributed Stochastic Neighbor Embedding (t-SNE) plots of feature space.
- Open Access
COVID-CT-H-UNet: a novel COVID-19 CT segmentation network based on attention mechanism and Bi-category Hybrid loss
Since 2019, the global COVID-19 outbreak has emerged as a crucial focus in healthcare research. Although RT-PCR stands as the primary method for COVID-19 detection, its extended detection time poses a significant challenge. Consequently, supplementing RT-PCR with the pathological study of COVID-19 through CT imaging has become imperative. The current segmentation approach based on TVLoss enhances the connectivity of afflicted areas. Nevertheless, it tends to misclassify normal pixels between certain adjacent diseased regions as diseased pixels. The typical Binary cross entropy(BCE) based U-shaped network only concentrates on the entire CT images without emphasizing on the affected regions, which results in hazy borders and low contrast in the projected output. In addition, the fraction of infected pixels in CT images is much less, which makes it a challenge for segmentation models to make accurate predictions. In this paper, we propose COVID-CT-H-UNet, a COVID-19 CT segmentation network to solve these problems. To recognize the unaffected pixels between neighbouring diseased regions, extra visual layer information is captured by combining the attention module on the skip connections with the proposed composite function Bi-category Hybrid Loss. The issue of hazy boundaries and poor contrast brought on by the BCE Loss in conventional techniques is resolved by utilizing the composite function Bi-category Hybrid Loss that concentrates on the pixels in the diseased area. The experiment shows when compared to the previous COVID-19 segmentation networks, the proposed COVID-CT-H-UNet's segmentation impact has greatly improved, and it may be used to identify and study clinical COVID-19.
Open AccessCOVID-CT-H-UNet: a novel COVID-19 CT segmentation network based on attention mechanism and Bi-category Hybrid loss
Since 2019, the global COVID-19 outbreak has emerged as a crucial focus in healthcare research. Although RT-PCR stands as the primary method for COVID-19 detection, its extended detection time poses a significant challenge. Consequently, supplementing RT-PCR with the pathological study of COVID-19 through CT imaging has become imperative. The current segmentation approach based on TVLoss enhances the connectivity of afflicted areas. Nevertheless, it tends to misclassify normal pixels between certain adjacent diseased regions as diseased pixels. The typical Binary cross entropy(BCE) based U-shaped network only concentrates on the entire CT images without emphasizing on the affected regions, which results in hazy borders and low contrast in the projected output. In addition, the fraction of infected pixels in CT images is much less, which makes it a challenge for segmentation models to make accurate predictions. In this paper, we propose COVID-CT-H-UNet, a COVID-19 CT segmentation network to solve these problems. To recognize the unaffected pixels between neighbouring diseased regions, extra visual layer information is captured by combining the attention module on the skip connections with the proposed composite function Bi-category Hybrid Loss. The issue of hazy boundaries and poor contrast brought on by the BCE Loss in conventional techniques is resolved by utilizing the composite function Bi-category Hybrid Loss that concentrates on the pixels in the diseased area. The experiment shows when compared to the previous COVID-19 segmentation networks, the proposed COVID-CT-H-UNet's segmentation impact has greatly improved, and it may be used to identify and study clinical COVID-19.