BRAC University Institutional Repository
Preserving Knowledge, Advancing Research, Sharing Scholarship
A digital platform for collecting, preserving, and sharing BRAC University’s scholarly, academic, and institutional outputs.
Contribute to the Repository
Faculty members and students are invited to submit their research publications, theses, dissertations, and scholarly works to increase visibility, access, and long-term preservation.
Recent Submissions
MIM-HeMIS: Self-supervised masked image modeling with heterogeneous modality fusion for brain tumor segmentation under missing MRI modalities
(BRAC University, 2026-06) Hasan, Md. Jahidul; Opy, Raduan Ahmed; Shafi, Md Alif Khan; Jameel, Ahnaf Ashraf; Alvi, Md. Jobayer; Ahmed, Md. Sabbir; Department of Computer Science and Engineering
Accurate segmentation of regions of interest from MRI to diagnose brain tumors is crucial. MRI is the standard technique used in imaging the brain, but manual segmentation of tumor regions is time-consuming, manual and prone to human error. Although deep learning has been able to significantly enhance segmentation
performance, it requires big data sets of labeled images which is difficult in medical
applications. In this study, we introduce a hybrid CNN-Transformer framework that
integrates the Masked Image Modeling (MIM) based self-supervised pretraining and
the HeMIS statistical abstraction mechanism. A hybrid model is pretrained on the
BraTS 2021 dataset and finetuned for 4-class brain tumor segmentation for all 4
MRI modalities following the proposed framework. A major advantage is that the
modalities are robust, meaning that there is no need to train a new model for each
of the 15 possible combinations of modalities. When modalities are missing, the
variance computed by the HeMIS abstraction layer captures feature disagreement
across available inputs, providing useful information for identifying cases where pre-
dictions may be less reliable. Though this variance has not been calibrated as a
clinical confidence score. The model was tested on BraTS 2021 with all 15 modality
combinations with excellent and clinically applicable results: Whole Tumor Dice of
90.2%, Tumor Core Dice of 85.9%, and Enhancing Tumor Dice of 79.1% for all four
modalities.
MemeFusionNet: A cross-linguistic multimodal model for identifying troll memes
(Institute of Electrical and Electronics Engineers Inc., 2025-01-01) Sultan T.; Akbarpour H.A.; El-Shafai W.; Saib M.; Bhuiyan, Md. Khairul Bashar; Islam M.S.; Azar A.T.; Njima C.B.; Department of Electrical and Electronic Engineering
The proliferation of troll memes, which exploit textual and visual elements to propagate misinformation and incite negativity, presents a critical challenge for online content moderation. Existing methods often struggle with cross-linguistic generalization, multimodal fusion, and contextual understanding, limiting their effectiveness in multilingual environments. To address these gaps, we propose MemeFusionNet, a transformer-driven multimodal fusion framework that effectively captures the intricate relationships between images and text. MemeFusionNet integrates a cross-modal attention mechanism based on ViLT to enhance contextual awareness and better detect implicit troll content, such as sarcasm and cultural nuances. Our model demonstrates superior performance on Bangla and English meme datasets, achieving 86% and 96% accuracy, respectively, outperforming all existing benchmarks. Its scalable architecture ensures robust cross-lingual adaptability, making it well-suited for large-scale, real-time content moderation.
Analysis of the correlation between SARS-CoV-2 transmission and meteorological parameters in Bangladesh
(SIPISS- Edizioni FS Publishers, 2022-01-01) Rahaman, Md. Mushfiqur; Khan, Risala T.; Zaman, Shakila; Rahman, Md. Tanvir; Shuva, Taslima F.; Department of Computer Science and Engineering
Introduction: Since COVID-19 has been characterized as a worldwide epidemic, multiple studies have suggested that weather may have a role in virus transmission. This research aims to examine the correlation between meteorological parameters and SARS-CoV-2 transmission, as well as to forecast cumulative COVID-19 cases in Bangladesh. Methods: In this study, an average incubation period of 5-6 days was used to examine the real effect of environmental parameters on SARS-CoV-2 transmission. Therefore, considering the incubation period and reporting time a standard 7-day shift in meteorological parameters from the daily COVID-19 cases was applied to measure the actual correlation. In this regard, the non-parametric correlation test (Spearman's Rank Correlation) was performed where 95% (p < 0.05) and 99% (p < 0.01) confidence intervals were considered as an acceptance criterion. Results: This work found a significant positive correlation (p < 0.01) for COVID-19 cases with minimum temperature, average temperature (only for division-specific analysis), wind speed, rainfall, humidity, and cloud. Furthermore, a significant negative correlation (p < 0.01) was found with atmospheric pressure and sun hours. However, the impact of maximum temperature (except for some divisions) or UV index was significantly low. Discussion: Our findings showed that the strength of the correlation coefficient is higher for the test positivity rate rather than the confirmed case count. However, to forecast the cumulative cases of COVID-19, ARIMA (Autoregressive Integrated Moving Average) may be considered the best-fitting model according to AIC (Akaike Information Criterion) and performs slightly better than Holt's exponential smoothing model. Additionally, this study represents the comparative analysis between predicted and actual Coronavirus-19 cases during December 2021 to show how close the predicted result is to the selected model. Take-home message: In this study, the overall analysis indicates that COVID-19 outbreaks in Bangladesh are more likely to hit massively during the pre-monsoon (March to May) season and the monsoon (June to October) season than in winter (November to February). These findings might be useful for decision-makers and authorities to know more about the seasonal impact of the outbreak and plan accordingly before the country enters a new weather season. © 2022 by the authors.
People with disabilities and transport access: Evidence from PENDA and inclusive futures1
(Institute of Development Studies, 2026-01-01) Carew, Mark T.; Das, Narayan; Thompson, Stephen; BRAC Institute of Governance and Development
Transport is a fundamental enabler of participation in society yet people with disabilities frequently experience barriers to accessing it. These barriers are often greater in low- and middle-income countries (LMICs). Despite its importance, transport remains neglected in disability-inclusive development interventions. This article synthesises evidence from the Disability Inclusive Development Inclusive Futures programme and the Programme for Evidence to Inform Disability Action, which implement and evaluate disability inclusion interventions across multiple LMICs. We draw on their findings to examine the transport barriers experienced by people with disabilities and how these shaped intervention participation and outcomes. Findings highlight inaccessible infrastructure, a lack of accessible public transport options, unavailability of assistive technology, and transport-related stigma and discrimination as barriers. These barriers were shaped by intersectional disadvantage and risk constraining intervention participation and impact. We discuss the implications for disability-inclusive development intervention design and national transport policy, highlighting that transport access is a prerequisite for meaningful inclusion. © 2026 The Authors. IDS Bulletin © Institute of Development Studies
PrecisionStroke: Optimized stroke risk prediction through advanced hyperparameter tuning and machine learning techniques
(Institute of Electrical and Electronics Engineers Inc., 2025-01-01) Arman, Mithila; Shuba S.J.; Rhaman M.M.; Choya S.B.Z.; Chowdhury M.A.A.; Islam M.; Sheikh I.A.; Jahan M.K.; Department of Computer Science and Engineering
Stroke prediction is a challenging problem in the healthcare domain, especially due to high class imbalance in datasets where data about stroke cases is a very small portion. In this study we develop a machine learning based framework for stroke prediction that tackles through powerful data preprocessing methods, such as Synthetic Minority Oversampling Technique (SMOTE) for balancing the classes. (Situated within the entire framework of a comparative evaluation of the Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), CatBoost, XGBoost, LightGBM, and Multi-Layer Perceptron (MLP) and TabNet machine learning classifiers as per unique designs. In conclusion, the Random Forest classifier proved to be the best performing model with an accuracy of 99.38%, followed closely by LightGBM and CatBoost. A detailed analysis including KP song metrics including precision, F1 score, and confusion matrices were carried out to ensure thorough evaluation. The study also emphasizes the importance of explainable AI tools which improve the explainability of predictions. This framework shows that machine learning has the potential to assist with early detection of stroke and may ultimately be integrated into clinical decision-making systems.