Mahir Afser Pavel

Dr. Mahir Afser Pavel

orcid

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
13 September 2024 in Drones
Verified
Journal Article
Open access
PDF file
Md Fahim Shahoriar TituMd Fahim Shahoriar Titu
Md Fahim Shahoriar Titu
Mahir Afser PavelMahir Afser Pavel
Mahir Afser Pavel
KGKah Ong Michael Goh
Kah Ong Michael Goh
HBHisham Babar
Hisham Babar
UAUmama Aman
Umama Aman
RKRiasat Khan
Riasat Khan
Fire accidents are life-threatening catastrophes leading to losses of life, financial damage, climate change, and ecological destruction. Promptly and efficiently detecting and extinguishing fires is essential to reduce the loss of lives and damage. This study uses drone, edge computing, and artificial intelligence (AI) techniques, presenting novel methods for real-time fire detection. This proposed work utilizes a comprehensive dataset of 7187 fire images and advanced deep learning models, e.g., Detection Transformer (DETR), Detectron2, You Only Look Once YOLOv8, and Autodistill-based knowledge distillation techniques to improve the model performance. The knowledge distillation approach has been implemented with the YOLOv8m (medium) as the teacher (base) model. The distilled (student) frameworks are developed employing the YOLOv8n (Nano) and DETR techniques. The YOLOv8n attains the best performance with 95.21% detection accuracy and 0.985 F1 score. A powerful hardware setup, including a Raspberry Pi 5 microcontroller, Pi camera module 3, and a DJI F450 custom-built drone, has been constructed. The distilled YOLOv8n model has been deployed in the proposed hardware setup for real-time fire identification. The YOLOv8n model achieves 89.23% accuracy and an approximate frame rate of 8 for the conducted live experiments. Integrating deep learning techniques with drone and edge devices demonstrates the proposed system’s effectiveness and potential for practical applications in fire hazard mitigation.

6 November 2024 in PLOS ONE
Verified
Journal Article
Open access
PDF file
Mahir Afser PavelMahir Afser Pavel
Mahir Afser Pavel
RIRafiul Islam
Rafiul Islam
SBShoyeb Bin Babor
Shoyeb Bin Babor
RMRiaz Mehadi
Riaz Mehadi
RKRiasat Khan
Riasat Khan
Non-small cell lung cancer (NSCLC) exhibits a comparatively slower rate of metastasis in contrast to small cell lung cancer, contributing to approximately 85% of the global patient population. In this work, leveraging CT scan images, we deploy a knowledge distillation technique within teaching assistant (TA) and student frameworks for NSCLC classification. We employed various deep learning models, CNN, VGG19, ResNet152v2, Swin, CCT, and ViT, and assigned roles as teacher, teaching assistant and student. Evaluation underscores exceptional model performance in performance metrics achieved via cost-sensitive learning and precise hyperparameter (alpha and temperature) fine-tuning, highlighting the model’s efficiency in lung cancer tumor prediction and classification. The applied TA (ResNet152) and student (CNN) models achieved 90.99% and 94.53% test accuracies, respectively, with optimal hyperparameters (alpha = 0.7 and temperature = 7). The implementation of the TA framework improves the overall performance of the student model. After obtaining Shapley values, explainable AI is applied with a partition explainer to check each class’s contribution, further enhancing the transparency of the implemented deep learning techniques. Finally, a web application designed to make it user-friendly and classify lung types in recently captured images. The execution of the three-stage knowledge distillation technique proved efficient with significantly reduced trainable parameters and training time applicable for memory-constrained edge devices.

15 April 2025 in IEEE Access
Verified
Journal Article
Open access
AAAfifa Zain Apurba
Afifa Zain Apurba
Md Fahim Shahoriar TituMd Fahim Shahoriar Titu
Md Fahim Shahoriar Titu
Mahir Afser PavelMahir Afser Pavel
Mahir Afser Pavel
INIntisar Naheen
Intisar Naheen
RKRiasat Khan
Riasat Khan
Root canal treatment involves the removal of inflamed pulp from infected teeth and sealing them to prevent re-entry of bacteria and infection. Early and accurate identification of root canal issues is crucial for improving treatment outcomes and minimizing complications. This study proposes a novel method for efficiently detecting failures in root canal treatments by incorporating image filtering techniques into deep learning models. A custom dataset has been collected from Square Hospital Ltd of Dhaka, Bangladesh, comprising 1,600 annotated, filtered radiographs. These images are processed using denoising algorithms (mean, median, Gaussian, contourlet transform, or Bayesian wavelet), non-local means, and BM3D. The YOLO models used for detection included YOLOv5, YOLOv7, and YOLOv8, with YOLOv5 achieving the highest accuracy of 90.6% on the filtered datasets. This work integrates an autodistillation technique with Grounded SAM as the base model for bounding box generation, YOLOv8 as an intermediate model to bridge architectural differences, and YOLOv5 as the target model due to its lightweight design, fewer parameters, and superior performance on the filtered dataset. Knowledge distillation was employed to optimize performance on lightweight models, resulting in precision and recall values of 99.16% and 97.29%, respectively, for YOLOv5. The filtered datasets outperformed raw images, with the Bayesian wavelet and contourlet transform yielding the best results on YOLOv5. This pipeline demonstrates the potential to develop one of the most accurate and clinically applicable diagnostic tools for root canal failure detection by combining deep learning and knowledge distillation techniques with image filtering. This study demonstrates how combining deep learning with advanced image enhancement techniques can provide a highly accurate, clinically relevant tool for the early detection of root canal treatment failures, potentially aiding dentists in making faster and more consistent diagnoses.

25 August 2025 in Vehicles
Verified
Journal Article
Open access
PDF file
ASAnanya Saha
Ananya Saha
Mahir Afser PavelMahir Afser Pavel
Mahir Afser Pavel
Md Fahim Shahoriar TituMd Fahim Shahoriar Titu
Md Fahim Shahoriar Titu
AAAfifa Zain Apurba
Afifa Zain Apurba
RKRiasat Khan
Riasat Khan
Efficient and explainable vehicle damage inspection is essential due to the increasing complexity and volume of vehicular incidents. Traditional manual inspection approaches are not time-effective, prone to human error, and lead to inefficiencies in insurance claims and repair workflows. Existing deep learning methods, such as CNNs, often struggle with generalization, require large annotated datasets, and lack interpretability. This study presents a robust and interpretable deep learning framework for vehicle damage classification, integrating Vision Transformers (ViTs) and ensemble detection strategies. The proposed architecture employs a RetinaNet backbone with a ViT-enhanced detection head, implemented in PyTorch using the Detectron2 object detection technique. It is pretrained on COCO weights and fine-tuned through focal loss and aggressive augmentation techniques to improve generalization under real-world damage variability. The proposed system applies the Weighted Box Fusion (WBF) ensemble strategy to refine detection outputs from multiple models, offering improved spatial precision. To ensure interpretability and transparency, we adopt numerous explainability techniques—Grad-CAM, Grad-CAM++, and SHAP—offering semantic and visual insights into model decisions. A custom vehicle damage dataset with 4500 images has been built, consisting of approximately 60% curated images collected through targeted web scraping and crawling covering various damage types (such as bumper dents, panel scratches, and frontal impacts), along with 40% COCO dataset images to support model generalization. Comparative evaluations show that Hybrid ViT-RetinaNet achieves superior performance with an F1-score of 84.6%, mAP of 87.2%, and 22 FPS inference speed. In an ablation analysis, WBF, augmentation, transfer learning, and focal loss significantly improve performance, with focal loss increasing F1 by 6.3% for underrepresented classes and COCO pretraining boosting mAP by 8.7%. Additional architectural comparisons demonstrate that our full hybrid configuration not only maintains competitive accuracy but also achieves up to 150 FPS, making it well suited for real-time use cases. Robustness tests under challenging conditions, including real-world visual disturbances (smoke, fire, motion blur, varying lighting, and occlusions) and artificial noise (Gaussian; salt-and-pepper), confirm the model’s generalization ability. This work contributes a scalable, explainable, and high-performance solution for real-world vehicle damage diagnostics.

23 June 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
Mahir Afser PavelMahir Afser Pavel
Mahir Afser Pavel
NFNafiz Fahad
Nafiz Fahad
MHMd. Tanzib Hosain
Md. Tanzib Hosain
MMMd. Kishor Morol
Md. Kishor Morol
TLTze Hui Liew
Tze Hui Liew
Background: Segmentation of brain tumor subregions from multi-modal 3D MRI is necessary for clinical diagnosis and treatment planning. However, it continues to pose significant challenges due to severe class imbalance, ambiguous boundaries, and fragmented enhancing tumor (ET) regions. While convolutional and transformer-based models improve local and global feature representations, most methods operate in dense voxel or token spaces, resulting in high computational cost and a lack of explicit structural representation.Methods: This paper proposes SLGRA-Net, a Structure-aware Latent Graph Reasoning Network that enables global structural reasoning in a compact latent graph space. The framework aggregates 3D feature patches into graph nodes, allowing the efficient capture of long-range structural dependencies while preserving voxel-level precision. Graph-based message passing refines node features, which are adaptively fused into the decoder via gated unpooling. To address severe class imbalance and label uncertainty, particularly for ET, we introduce an auxiliary node-level supervision mechanism that is uncertainty-aware and employs class-balanced weighting.Findings: Extensive experiments on the BraTS benchmarks demonstrate that SLGRA-Net achieves strong performance (0.89 Dice), particularly for the clinically critical ET region, while significantly reducing computational overhead compared to transformer-based methods. Comprehensive ablation studies validate each architectural component, and external validation confirms the framework’s robust and consistent performance across diverse datasets.

18 June 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
Mahir Afser PavelMahir Afser Pavel
Mahir Afser Pavel
NFNafiz Fahad
Nafiz Fahad
MMMd. Kishor Morol
Md. Kishor Morol
TLTze Hui Liew
Tze Hui Liew
MHMd. Tanzib Hosain
Md. Tanzib Hosain
Brain tumors remain a major challenge in neuro-oncology due to diagnostic complexity, inter-observer variability, and increasing demand for reliable and privacy-aware computational analysis. Although deep learning has improved imaging-based diagnosis, practical deployment remains constrained by data decentralization, robustness of representations, class imbalance, and limited interpretability. This study presents a federated multi-component framework for brain tumor diagnosis that combines privacy-preserving federated optimization, generative augmentation, dual-view representation learning, contrastive feature learning, and explainable artificial intelligence. Rather than introducing a new learning paradigm, the framework is designed to investigate how these complementary components interact to improve representation quality, robustness, and interpretability under decentralized training settings. Experimental evaluation on the Brain MRI ND-5 dataset demonstrated classification performance exceeding 96\% across Accuracy and F1-score. Component-wise ablation studies and comparative analyses indicated that the integrated configuration consistently improved performance relative to individual modules and baseline architectures while maintaining privacy-aware distributed training and interpretable decision support. The findings suggest that combining federated learning, generative augmentation, contrastive representation learning, and explainability can provide a practical and scalable approach for MRI-based brain tumor diagnosis. Future work will focus on broader multi-institutional validation, multimodal integration, efficient training strategies, and further assessment of clinical usability and generalization.