Mostafa Karami

Mr. Mostafa Karami

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
11 November 2025
Verified
Book Chapter
Closed access
PDF file
JMJosué Martínez-Martínez
Josué Martínez-Martínez
OBOlivia Brown
Olivia Brown
MKMostafa Karami
Mostafa Karami
SNSheida Nabavi
Sheida Nabavi

22 November 2024
Verified
Conference Paper
Open access
SHSahand Hamzehei
Sahand Hamzehei
GRGianna Raimondi
Gianna Raimondi
MKMostafa Karami
Mostafa Karami
LOLinnaea Ostroff
Linnaea Ostroff
SNSheida Nabavi
Sheida Nabavi
Image registration is important in biological image analysis; however, it is often challenged by distortions and non-linear transformations. In this paper, we present a novel patch-wise image registration method to address the mentioned issues. Our method begins with global registration to correct linear transformations, followed by a detailed examination of geometrical distortions. After that, each image is adaptively divided into patches to isolate and correct non-linear distortions, followed by reconstruction and combining patches using Otsu thresholding. We evaluated our method against state-of-the-art techniques using mutual information (MI), phase congruency-based (PCB), and gradient-based metrics (GBM) across four real biology datasets. Our results demonstrate superior feature alignment and image coherence, especially in serial-stack registrations. While the proposed method has longer processing times compared to linear registration methods, its enhanced accuracy and reliability to handle non-uniform distortion makes it beneficial for precision-demanding applications. We have created a public GitHub repository containing the code used in our research, available at https://github.com/NabaviLab/CAPTURE.

20 June 2025 in arXiv
Verified
Preprint
Open access
PDF file
JMJosué Martínez-Martínez
Josué Martínez-Martínez
OBOlivia Brown
Olivia Brown
MKMostafa Karami
Mostafa Karami
SNSheida Nabavi
Sheida Nabavi
Deep neural networks are increasingly being used to detect and diagnose medical conditions using medical imaging. Despite their utility, these models are highly vulnerable to adversarial attacks and distribution shifts, which can affect diagnostic reliability and undermine trust among healthcare professionals. In this study, we propose a robust training algorithm with data augmentation (RTDA) to mitigate these vulnerabilities in medical image classification. We benchmark classifier robustness against adversarial perturbations and natural variations of RTDA and six competing baseline techniques, including adversarial training and data augmentation approaches in isolation and combination, using experimental data sets with three different imaging technologies (mammograms, X-rays, and ultrasound). We demonstrate that RTDA achieves superior robustness against adversarial attacks and improved generalization performance in the presence of distribution shift in each image classification task while maintaining high clean accuracy.

22 November 2024
Verified
Conference Paper
Open access
MKMostafa Karami
Mostafa Karami
BLBingjun Li
Bingjun Li
SWSamson Weiner
Samson Weiner
SHSahand Hamzehei
Sahand Hamzehei
SNSheida Nabavi
Sheida Nabavi
Detecting copy number variations (CNVs) in single-cell DNA sequencing (scDNA-seq) data is challenging due to substantial noise and variability. To address this, we present DCCNV, a novel method that integrates diffusion processes, contrastive learning, and circular binary segmentation (CBS) for reliable CNV detection. Our method employs adaptive k-nearest neighbors (KNN) and multi-scale diffusion to reduce noise while preserving key biological signals, followed by contrastive learning to distinguish true genomic alterations from technical noise. The CBS algorithm is then used to partition the enhanced signals into discrete copy number segments. We compared the performance of DCCNV with those of several current single-cell CNV detection methods, including DeepCopy, rcCAE, SCOPE, SCONE, HMMcopy, SeCNV, as well as filtering-based CNV detection approaches that employ commonly used filters such as Wavelet, Median, and Gaussian filters. This comparison was conducted using both simulated and real data. The results show that DCCNV outperforms these approaches in terms of accuracy and computational efficiency. The code used in this research is publicly available at https://github.com/NabaviLab/DCCNV.

12 February 2025 in arXiv
Verified
Preprint
Open access
PDF file
AMArman Mohammadigilani
Arman Mohammadigilani
HAHani Attar
Hani Attar
HCHamidreza Ehsani Chimeh
Hamidreza Ehsani Chimeh
MKMostafa Karami
Mostafa Karami
Parkinson's disease (PD) is a neurodegenerative condition characterized by notable motor and non-motor manifestations. The assessment tool known as the Unified Parkinson's Disease Rating Scale (UPDRS) plays a crucial role in evaluating the extent of symptomatology associated with Parkinson's Disease (PD). This research presents a complete approach for predicting UPDRS scores using sophisticated Long Short-Term Memory (LSTM) networks that are improved using attention mechanisms, data augmentation techniques, and robust feature selection. The data utilized in this work was obtained from the UC Irvine Machine Learning repository. It encompasses a range of speech metrics collected from patients in the early stages of Parkinson's disease. Recursive Feature Elimination (RFE) was utilized to achieve efficient feature selection, while the application of jittering enhanced the dataset. The Long Short-Term Memory (LSTM) network was carefully crafted to capture temporal fluctuations within the dataset effectively. Additionally, it was enhanced by integrating an attention mechanism, which enhances the network's ability to recognize sequence importance. The methodology that has been described presents a potentially practical approach for conducting a more precise and individualized analysis of medical data related to Parkinson's disease.

30 June 2026
Verified
Conference Paper
Open access
SHSahand Hamzehei
Sahand Hamzehei
MKMostafa Karami
Mostafa Karami
AJAfsana Ahsan Jeny
Afsana Ahsan Jeny
SBStephen Andrew Baker
Stephen Andrew Baker
TRTucker Van Rathe
Tucker Van Rathe
SNSheida Nabavi
Sheida Nabavi
Longitudinal comparison of prior and current mammograms is fundamental to breast cancer assessment. However, many deep learning approaches still rely on similarity measures grounded in flat Euclidean geometry, which may fail to capture correlated structural changes in feature representations. In this work, we propose a quantum-information-inspired distance function for Siamese networks in longitudinal mammogram imaging. Rather than computing similarity directly between feature vectors, we reinterpret embeddings as normalized statistical state representations and measure their distinguishability using a fidelity-based spectral distance. This formulation captures second-order feature correlations and nonlinear structure, providing a principled generalization of classical kernel-based distances. Our proposed distance is integrated into a longitudinal pipeline consisting of backbone feature extraction, quantum-inspired similarity computation, and binary classification. Our experiments across multiple recent Siamese and transformer-based models demonstrate consistent improvements in accuracy and related classification metrics when replacing conventional distance functions. The improvements remain across multiple backbones and under cross-dataset validation, indicating robustness and preliminary cross-dataset consistency. These results suggest that modeling longitudinal similarity as a state distinguishability problem offers a principled geometric foundation for improving deep learning-based mammographic analysis. All codes used in this paper are available at https://github.com/NabaviLab/Quantum-Information-Inspired-Distance-Functions-.git