Yasir Arafat Prodhan

Vision-language models (VLMs) have recently shown excellent progress in open-ended image-to-text generation. However, their multimodal nature makes them persistently vulnerable to backdoor attacks. Existing backdoor triggers for VLMs are either spatial, textual, or bimodal, which may yield localized or recognizable trigger patterns. In this work, we explore a different attack surface and propose \ textsc {FreqDoor}, a training-time backdoor attack that implants triggers in the frequency domain. \ textsc {FreqDoor} mixes amplitude-spectrum components from a trigger-source image selectively while preserving the phase of a clean image to generate a spatially distributed and visually imperceptible trigger without modifying the textual input. We evaluate the attack on BLIP-2, InstructBLIP, and LLaVA for image captioning and visual question answering. On Flickr8k, \ textsc {FreqDoor} achieves attack success rates of 99.6%, 99.8%, and 98.4% on the three models, respectively, while preserving the semantic quality of the generated captions. On VQAv2, the corresponding attack success rates are 99.6%, 92.4%, and 79.6%.

Fully sampled MRI acquisitions require dense traversal of k-space, resulting in inherently long scan times that limit clinical throughput and increase susceptibility to patient motion. A common strategy to accelerate MRI is to acquire only a subset of k-space measurements and reconstruct the missing information computationally. However, direct reconstruction from undersampled k-space is a severely ill-posed inverse problem that introduces aliasing artifacts, noise amplification, and loss of fine anatomical details. Conventional methods such as Parallel Imaging (PI) and Compressed Sensing (CS) partially address these challenges, while recent deep learning approaches have significantly improved reconstruction quality. Nevertheless, preserving high-frequency structures and maintaining robustness under aggressive undersampling remain challenging for existing convolution-based variational network frameworks. In this work, we propose a Variational Network with Wavelet-based U-Net (W-UNet) for accelerated MRI reconstruction from undersampled k-space data. The proposed framework combines physics-guided iterative reconstruction with learnable multi-scale frequency representations. Specifically, standard pooling operations are replaced by Discrete Wavelet Transform (DWT) and Inverse Wavelet Transform (IDWT) modules, enabling lossless downsampling while retaining both low-frequency structural information and high-frequency edge details. Integrated within the refinement and sensitivity map estimation stages of a variational network, the proposed design improves artifact suppression, feature preservation, and reconstruction fidelity for both single-coil and multi-coil settings. Our method achieves state-of-the-art results on the fastMRI knee and M4Raw brain datasets. We further validate the effectiveness of wavelet-based feature decomposition through comprehensive ablation studies, confirming its contribution to accelerated MRI reconstruction.

Retinopathy of Prematurity (ROP) is disease affecting mostly preterm babies with low birth-weights. Being the leading cause of vision loss in case of children worldwide \cite{c1}, it even caused epidemics in the 20th century \cite{c2}. Since it shows no early symptoms \cite{b23}, it becomes tough to diagnose ROP at an early stage. The stages of ROP depend on the level of the abnormal growth of the blood vessels of the retina. More importantly, detecting the stage of ROP is very difficult as there is no clear demarcation and the level of detachment or disruption of retina is hard to comprehend seeing fundus/retinal images. In the recent years, deep learning models and algorithms have proven their potentiality in early detection of several diseases and classification of stages based on medical images

Image segmentation is an important segment of medical image diagnosis. Due to the recent advancement in the field of deep learning, image segmentation can be possible without any human involvement. Applications of CNN and transformer based models are being used in every field nowadays. CNNs based models can learn the local features efficiently whereas transformer based models can learn the global features which makes combining these two a great tool for medical image segmentation. Again, converting Cartesian image to polar image reduces the dimension of the elliptical and complex decision boundaries of medical images to a more simple one. In this study, we will try to compare the scores of CNN and transformer based models on the dataset ISIC-2018 for medical image semantic segmentation. For the CNN based model, we will use UNET architecture. We will also use TransUNET which leverages the benefits of both CNNs and transformer. We will also compare the scores obtained from both Cartesian images and polar images. Index Terms—Convolution neural network, medical image processing, medical image segmentation, semantic segmentation, TransUnet, Stacked Hourglass