Pixel by Pixel Semantic Segmentation Approach on WSI Images for Gastric Gland Segmentation and Gastric Cancer Grade Classification Using MLP-XAI Model Mousumi Gupta, Prasanna Dhungel, Madhab Nirola, Bidyut Krishna Goswami, Amlan Gupta International Journal of Imaging Systems and Technology, 2025 Gastric cancer remains one of the most prevailing cancers with high mortality. Timely and quantitative diagnosis stays challenging with the pathologists. H&E stain provides a color composition which distinguishes individual components of gastric histopathology images. The human eye is able to distinguish each component but fails to quantify and varies with the pathologists' opinions. The gastric histopathology components like lamina propria sometimes contain hyperchromatic nuclei and lymphocytes. This characteristic sometimes makes the diagnosis confusing as the system might incorrectly identify it as malignancy. Automation of this diagnosis is extremely crucial but can be a strong support system in gastric cancer diagnosis. This study developed a combinational neural network approach based on DeepLabV3+ and U‐Net architectures. A pixel‐by‐pixel semantic segmentation approach is implemented to segment gland texture from gastric histopathology WSI images. A sliding window approach is employed to process the whole slide images. Various categories of gastric abnormalities classification models are implemented using Multilayer Perceptron (MLP). To interpret the classification model, the XAI technique is used, utilizing SHapley Additive exPlanations (SHAP). The model is able to categorize gastric lesions into five classes: benign, mild dysplasia, dysplasia, high‐grade dysplasia, and malignant using the features nuclear‐cytoplasmic ratio, GLCM, and intensity metrics. The segmentation model scored an accuracy of 96.983%, precision of 94.057%, recall of 93.835%, and F1 score of 95.497%, and the classification model achieved an accuracy of 90.36%. A framework is designed to support pathologists in making early decisions on gastric cancer.
Drug utilisation patterns & clinical outcomes in hospitalised COVID-19 patients: A geospatial & machine learning approach Dhruva Kumar Sharma, Madhab Nirola, Mousumi Gupta, Arpan Sharma, Prasanna Dhungel, Barun Kumar Sharma Indian Journal of Medical Research, 2025 Background & objectives Coronavirus disease (COVID-19) caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has posed challenges in clinical management due to a lack of established treatment guidelines. This study aimed to analyse drug utilisation patterns and identify factors influencing clinical outcomes in COVID-19 patients. Methods A retrospective study was conducted on 380 confirmed COVID-19 patients admitted between April and June 2021 at a tertiary hospital in Sikkim, India. Study participants demographics, medications, comorbidities, outcomes, and geospatial data were collected with due approval from the Institutional Ethics Committee. Machine learning classification and regression models were used for analysis. Results The Random Forest classification model achieved the highest accuracy of 90.7 per cent and an AUROC score of 0.86. Methylprednisolone use was associated with an 11.4 per cent mortality rate. Geospatial analysis identified significant mortality clustering in the East district for female study participants and in the East and North districts for male study participants, with a Moran’s I index of 0.125080 and a z-score of 8.642819, indicating statistically significant spatial clustering. Interpretation & conclusions The study provides insights into COVID-19 management practices and outcomes. Machine learning identified relationships between factors associated with mortality, which could be due to advanced disease state, associated co-morbidities or post-treatment issues. Further prospective studies are needed to validate findings and address limitations.
Critical perspectives on climate change and Glacial Lake outburst floods' impact in the Himalayas: Policy inferences from the South Lhonak Flood of 2023 Arpan Sharma, Narpati Sharma, Mousumi Gupta Water Resources Management in Mountain Regions, 2025 The Himalayan region is facing a pressing environmental crisis marked by the rapid expansion of glacial lakes. This surge, primarily fueled by climate change and compounded by seismic activity and landslides, presents a severe threat to communities downstream. This chapter delves into the issue of glacial lake expansion, focusing particularly on glacial lake outburst floods (GLOFs). Through a case study of the 2023 South Lhonak GLOF, which caused significant casualties and damage, the study extracts crucial insights for policy formulation. It emphasizes the need for a collaborative policy approach to tackle the expanding glacial lake problem. Drawing from past disasters, we underscore the increasing frequency and severity of GLOFs, supported by recent empirical studies, and examine the impact of the South Lhonak Lake GLOF on the Teesta River valley. This event underscores the urgency for policymakers to reassess existing risk assessment and disaster management frameworks. The chapter identifies the shortcomings in current glacial lake monitoring in the Himalayas and the lack of robust early warning systems as critical deficiencies that demand immediate attention. The primary objective of this chapter is to guide policy decisions aimed at reducing the vulnerability of Himalayan communities to GLOF incidents and fostering resilience in the face of environmental change.
Enhanced Feature Based Methods for Histopathology Image Registration with ORB Bijoyeta Roy, Mousumi Gupta 2025 IEEE International Conference on Electronics Computing and Communication Technologies Conecct 2025, 2025 Histopathology image registration is a critical process for accurate diagnosis and analysis in medical imaging. Feature-based methods, compared to intensity-based or transform-based techniques, are particularly effective in aligning histopathology images due to their robustness against variations in staining, lighting, and tissue deformation. These methods allow for precise comparison and interpretation of tissue samples, enhancing diagnostic accuracy. In this study, we employ the ORB (Oriented FAST and Rotated BRIEF) detector to generate key points and use the BRIEF (Binary Robust Independent Elementary Features) algorithm for feature matching. The ORB detector is advantageous due to its computational efficiency and invariance to rotation and scale, making it ideal for histopathological applications. To further enhance performance, adaptive thresholding is incorporated in ORB for improved feature detection in varying intensity regions. Additionally, RANSAC (Random Sample Consensus) is applied for robust outlier rejection, ensuring accurate alignment of histopathology images. This combination has demonstrated successful results in aligning histopathology images, ensuring high accuracy and consistency. The proposed method effectively addresses challenges posed by variations in tissue appearance, contributing to improved diagnostic precision and research in medical imaging.
Integration of UNET and GAN for specular reflection detection and inpainting in colposcopy medical images Parimala Tamang, Annet Thatal, Mousumi Gupta, Nishank Upreti 2025 IEEE International Conference on Electronics Computing and Communication Technologies Conecct 2025, 2025 The colposcopy procedure is a gold standard for the detection of cervical neoplasia which uses a device called colposcope. Colposcope produces a digital image which often comes with a glare also known as specular reflection. Specular reflections in colposcopy images are considered artifacts which hide the underlying tissues in colposcopy images. Furthermore, during colposcopy image processing and analysis, the presence of specular reflections can interfere with subsequent image processing tasks. This study presents an integration of UNET and generative network architecture for the detection and inpainting of specular regions in colposcopy images in two stages. The first stage focuses on the detection of specular reflection regions in 2778 colposcopy images obtained from the International Agency for Research in Cancer. The second stage focuses on the inpainting of specular reflection regions in the colposcopy dataset. Subjective evaluation is performed to assess the model’s performance by an expert. Furthermore, performance metrics SSIM scored 0.883, showing a good performance by the model in effective detection and inpainting of specular reflection regions.
Tracing the COVID-19 spread pattern in India through a GIS-based spatio-temporal analysis of interconnected clusters Mousumi Gupta, Arpan Sharma, Dhruva Kumar Sharma, Madhab Nirola, Prasanna Dhungel, Ashok Patel, Harpreet Singh, Amlan Gupta Scientific Reports, 2024 Spatiotemporal analysis is a critical tool for understanding COVID-19 spread. This study examines the pattern of spatial distribution of COVID-19 cases across India, based on data provided by the Indian Council of Medical Research (ICMR). The research investigates temporal patterns during the first, second, and third waves in India for an informed policy response in case of any present or future pandemics. Given the colossal size of the dataset encompassing the entire nation’s data during the pandemic, a time-bound convenience sampling approach was employed. This approach was carefully designed to ensure a representative sample from advancing timeframes to observe time-based patterns in data. Data were captured from March 2020 to December 2022, with a 5-day interval considered for downloading the data. We employ robust spatial analysis techniques, including the Moran’s I index for spatial correlation assessment and the Getis Ord Gi* statistic for cluster identification. It was observed that positive COVID-19 cases in India showed a positive auto-correlation from May 2020 till December 2022. Moran’s I index values ranged from 0.11 to 0.39. It signifies a strong trend over the last 3 years with $$r^2$$ r 2 of 0.74 on order 3 polynomial regression. It is expected that high-risk zones can have a higher number of cases in future COVID-19 waves. Monthly clusters of positive cases were mapped through ArcGIS software. Through cluster maps, high-risk zones were identified namely Kerala, Maharashtra, New Delhi, Tamil Nadu, and Gujarat. The observation is: high-risk zones mostly fall near coastal areas and hotter climatic zones, contrary to the cold Himalayan region with Montanne climate zone. Our aggregate analysis of 3 years of COVID-19 cases suggests significant patterns of interconnectedness between the Indian Railway network, climatic zones, and geographical location with COVID-19 spread. This study thereby underscores the vital role of spatiotemporal analysis in predicting and managing future COVID-19 waves as well as future pandemics for an informed policy response.
Revolutionizing Colon Histopathology Glandular Segmentation Using an Ensemble Network With Watershed Algorithm Bijoyeta Roy, Mousumi Gupta, Bidyut Krishna Goswami International Journal of Imaging Systems and Technology, 2024 Colorectal adenocarcinoma, the most prevalent form of colon cancer, originates in the glandular structures of the intestines, presenting histopathological abnormalities in affected tissues. Accurate gland segmentation is crucial for identifying these potentially fatal abnormalities. While recent methodologies have shown success in segmenting glands in benign tissues, their efficacy diminishes when applied to malignant tissue segmentation. This study aims to develop a robust learning algorithm using a convolutional neural network (CNN) to segment glandular structures in colon histology images. The methodology employs a CNN based on the U‐Net architecture, augmented by a weighted ensemble network that integrates DenseNet 169, Inception V3, and Efficientnet B3 as backbone models. Additionally, the segmented gland boundaries are refined using the watershed algorithm. Evaluation on the Warwick‐QU dataset demonstrates promising results for the ensemble model, by achieving an F1 score of 0.928 and 0.913, object dice coefficient of 0.923 and 0.911, and Hausdorff distances of 38.97 and 33.76 on test sets A and B, respectively. These results are compared with outcomes from the GlaS challenge (MICCAI 2015) and existing research findings. Furthermore, our model is validated with a publicly available dataset named LC25000, and visual inspection reveals promising results, further validating the efficacy of our approach. The proposed ensemble methodology underscores the advantages of amalgamating diverse models, highlighting the potential of ensemble techniques to enhance segmentation tasks beyond individual model capabilities.
Automated Nuclei Analysis from Digital Histopathology Bijoyeta Roy, Pratima Sarkar, Mousumi Gupta Proceedings of 2023 International Conference on Intelligent Systems Advanced Computing and Communication Isacc 2023, 2023
