ZAHRA FASKA

ZAHRA FASKA

orcid

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

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

Representative Publications

View All Publications
7 July 2023 in Journal of Engineering and Applied Science
Verified
Journal Article
Open access
PDF file
ZFZAHRA FASKA
ZAHRA FASKA
LKLahbib Khrissi
Lahbib Khrissi
KHKhalid Haddouch
Khalid Haddouch
NANabil El Akkad
Nabil El Akkad
In the present study, we aim to propose an effective and robust ensemble-learning approach with stacked generalization for image segmentation. Initially, the input images are processed for feature extraction and edge detection using the Gabor filter and the Canny algorithms, respectively; our main goal is to determine the most feature descriptions. Subsequently, we applied the stacking generalization technique, which is generally built with two main learning levels. The first level is composed of two algorithms that give good results in the literature, namely: LightGBM (Light Gradient Boosting Machine) and SVM (support vector machine). The second level is the meta-model in which we use a predictor model that takes the base-level predictions to improve the accuracy of the final prediction. In the stacked generalization process, we use the Extreme Gradient Boosting (XGBoost); it takes as input the sub-models’ outputs to better classify each pixel of the image to give the final prediction. Today, several research works exist in the literature using different machine learning algorithms; in fact, instead of trying to find a single efficient and optimal learner, ensemble-based techniques take the advantage of each basic model; they integrate their outputs to obtain a more consistent and reliable learner. The result obtained from the models of individuals and our proposed approach is compared using a set of evaluation measures for image quality such as IoU, DSC, CC, SSIM, SAM, and UQI. The evaluation and a comparison of the results obtained showed more consistent predictions for the proposed model. Thus, we have made a comparison with some recent deep learning-based unsupervised segmentation methods. The evaluation and a comparison of the results obtained showed more coherent predictions for our stacked generalization in terms of precision, robustness, and consistency.

28 May 2025 in IEEE Access
Verified
Journal Article
Open access
ZFZAHRA FASKA
ZAHRA FASKA
LKLahbib Khrissi
Lahbib Khrissi
IMImadeddine Mountasser
Imadeddine Mountasser
KHKhalid Haddouch
Khalid Haddouch
NANabil El Akkad
Nabil El Akkad
WEWalid El-Shafai
Walid El-Shafai
Increasingly available ultrastructural data from a continuously growing diversity of experimental conditions are driving new opportunities for fruitful neuroscientific hypotheses tested in intracellular compartments such as the nanoscale roles of, e.g., the mitochondria. Reliable morphological statistics are based on achieving highly accurate semantic segmentations of EM images. The state-of-the-art deep CNNs can be somewhat brittle; they tend to provide coarse and high-frequency-oscillatory solutions with discontinuities and false positives even for simple mitochondria segmentation. Historically, the current state-of-the-art in medical image segmentation would involve some variant of the encoder-decoder architecture, such as the U-Net architecture. The SAM does not perform as well, since it has not been explicitly trained for the task and does not demonstrate user-interactive, over one billion annotations mostly for natural images. However, the SAM may be applied to segment anything, including medical image segmentation challenging new datasets. This work is aimed at the difficult task of implementing domain adaptation in mitochondria segmentation within EM images obtained from various tissues and species, using deep learning. We do a systematic study to assess SAM’s ability to perform segmentation in medical images, measure its performance on volumetric EM datasets, and show that it is powerful at segmenting instances even under challenging imaging conditions. We provide a fine-tuning SAM which can be naturally trained by SAM at an exemplary scale, benefiting from a diverse and large dataset over one million image masks in 11 modalities. This model would be able to perform precise segmentation for a wide range of targets under various imaging conditions, at the level of performance of specialized U-Net models, or even better. A visual comparison is shown between our fine-tuning SAM model and U-Net, along with an examination of different watershed post-processing strategies to discriminate between adjacent or conjoined instances. Results from our experiments show that the method suggested is fast, very robust, and accurate, with an imbibed model that has improved learning capability. A comprehensive sensitivity analysis has been carried out, and an ablation study using most popular metrics segmentation evaluation is performed, the results quantified by Dice Similarity Coefficient, Jaccard-Index coefficient, Aggregated Jaccard-Index, Panoptic quality, which confirm the robustness and effectiveness of the introduced modules in enhancing the performance of mitochondria segmentation.

26 June 2021
Verified
Book Chapter
Closed access
PDF file
ZFZAHRA FASKA
ZAHRA FASKA
LKLahbib Khrissi
Lahbib Khrissi
KHKhalid Haddouch
Khalid Haddouch
NANabil EL Akkad
Nabil EL Akkad

28 April 2023
Verified
Book Chapter
Closed access
PDF file
ZFZAHRA FASKA
ZAHRA FASKA
LKLahbib Khrissi
Lahbib Khrissi
KHKhalid Haddouch
Khalid Haddouch
NANabil El Akkad
Nabil El Akkad

26 February 2026 in Multimedia Tools and Applications
Verified
Journal Article
Closed access
PDF file
ZFZAHRA FASKA
ZAHRA FASKA
LKLahbib Khrissi
Lahbib Khrissi
KHKhalid Haddouch
Khalid Haddouch
NANabil El Akkad
Nabil El Akkad