Sally Saad Ismail
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Sally Saad Ismail has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Esraa Abdelhamid
Ainshams University
4 shared publications - Mostafa Aref
Department Computer Science, Faculty of Computer and Information Sciences,Ain Shams University, Cairo, Egypt
4 shared publications - Mohamed Ahmed Abo El-Enen
Faculty of Computer and Information Sciences, Ain Shams University, Cairo 11566, Egypt
2 shared publications - Taymoor Nazmy
Computer Science Department, Faculty of Computer and Information Sciences, Ain Shams University, Cairo, Egypt
2 shared publications - Abdel-Badeeh M. Salem
Computer Science Department, Faculty of Computer and Information Science, Ain Shams University, Cairo 11566, Egypt
1 shared publication - Abeer Mahmoud Mahmoud
Computer Science Department, Faculty of Computer and Information Science, Ain Shams University, Cairo 11566, Egypt
1 shared publication
Latest Publications
- Open Access
- 6 October 2026
- Published by MDPI AG in Journal of Imaging
- Vol. 12 (10) , 491
Adaptive Decision-Level Fusion for Osteosarcoma Classification Using Unpaired X-Ray and Histopathology Data
Osteosarcoma is an aggressive bone malignancy in children and adolescents, where early and accurate classification is essential for improving clinical outcomes. While deep learning methods have shown promising results in medical image analysis, most existing approaches rely on single-modality data or assume patient-level correspondence in multimodal learning, which is rarely available in real-world clinical settings. In addition, many fusion strategies are designed for paired datasets and depend on fixed weighting schemes, limiting their applicability to heterogeneous and independently collected data. To address these challenges, this study proposes a modality-agnostic ensemble learning framework across independent cohorts for osteosarcoma classification using histopathology and X-ray images. Each modality is first modeled independently using EfficientNet-B0, and decision-level integration is performed without requiring cross-modal alignment. Two fusion strategies are investigated: fixed-weight probability fusion and a proposed adaptive sample-level gating network that dynamically learns fusion weights based on modality-specific predictive confidence. Experiments conducted on publicly available histopathology (TCIA) and X-ray (BTXRD) datasets demonstrate that the proposed adaptive fusion approach achieves superior performance compared to fixed fusion. The adaptive gating network reaches an accuracy of 97.08%, outperforming fixed-weight fusion (94.15%) and unimodal baselines. These results confirm that adaptive decision-level fusion can effectively leverage complementary information across unpaired datasets. Overall, the proposed framework provides a practical solution for multimodal medical image classification under data fragmentation conditions, where paired datasets are not available.
- 6 October 2026
- Published by MDPI AG in Journal of Imaging
- Vol. 12 (10) , 491
Open AccessAdaptive Decision-Level Fusion for Osteosarcoma Classification Using Unpaired X-Ray and Histopathology Data
Osteosarcoma is an aggressive bone malignancy in children and adolescents, where early and accurate classification is essential for improving clinical outcomes. While deep learning methods have shown promising results in medical image analysis, most existing approaches rely on single-modality data or assume patient-level correspondence in multimodal learning, which is rarely available in real-world clinical settings. In addition, many fusion strategies are designed for paired datasets and depend on fixed weighting schemes, limiting their applicability to heterogeneous and independently collected data. To address these challenges, this study proposes a modality-agnostic ensemble learning framework across independent cohorts for osteosarcoma classification using histopathology and X-ray images. Each modality is first modeled independently using EfficientNet-B0, and decision-level integration is performed without requiring cross-modal alignment. Two fusion strategies are investigated: fixed-weight probability fusion and a proposed adaptive sample-level gating network that dynamically learns fusion weights based on modality-specific predictive confidence. Experiments conducted on publicly available histopathology (TCIA) and X-ray (BTXRD) datasets demonstrate that the proposed adaptive fusion approach achieves superior performance compared to fixed fusion. The adaptive gating network reaches an accuracy of 97.08%, outperforming fixed-weight fusion (94.15%) and unimodal baselines. These results confirm that adaptive decision-level fusion can effectively leverage complementary information across unpaired datasets. Overall, the proposed framework provides a practical solution for multimodal medical image classification under data fragmentation conditions, where paired datasets are not available.
- Open Access
- 17 June 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (12) , 6158
Med-LLaMA3: Advancing Medical Question-Answering Through Parameter-Efficient Fine-Tuning of Large Language Models
Despite recent advances, medical question answering systems still struggle with domain-specific reasoning and data efficiency. This paper presents Med-LLaMA3, a family of medical large language models developed by parameter-efficient fine-tuning of the LLaMA-3.1 (8 billion) and LLaMA-3.2 (1 and 3 billion) architectures using quantized low-rank adaptation (QLoRA) and low-rank adaptation (LoRA) with 4-bit quantization. Beyond model training, this work contributes the following: (1) a formalized dataset curation taxonomy (source type × clinical granularity × task format) with a source-category ablation confirming that the multi-source combination drives benchmark gains beyond any single category; (2) a systematic characterization of low-rank-adaptation rank-scaling behavior for the LLaMA-3 family in the medical domain (monotonic improvement up to rank 128, with no observed plateau); and (3) statistically validated comparisons using McNemar’s test and 95% bootstrap confidence intervals. We curated a medical instruction dataset of over 1.5 million samples spanning medical examinations, clinical dialogues, and biomedical literature. Our approach trains only ∼4% of the base model’s parameters and, consistent with prior studies of parameter-efficient methods in the medical domain, achieves performance comparable to full fine-tuning at a fraction of the memory footprint. Evaluated with five in-context examples per prompt, the 8-billion-parameter model attains a mean accuracy of 75.71% across the eight medical-domain subsets of the Massive Multitask Language Understanding benchmark; improvements over the unmodified LLaMA-3.1-8B-Instruct baseline are statistically significant on the medical multiple-choice benchmark MedMCQA and, after Bonferroni correction across the eight subsets, on three subsets (Clinical Knowledge, Medical Genetics, and Nutrition), with two further subsets being significant only before correction. A structured named-entity-recognition evaluation on 100 hospital discharge summaries (macro-averaged F1 0.94; dual-annotator agreement ) provides complementary evidence of clinical-text utility. A safety mitigation pilot shows that context-disambiguation preprocessing reduces the highest-severity abbreviation-ambiguity error rate from 30% to 10% on a 30-case held-out set. These results show that parameter-efficient fine-tuning can deliver high-performance medical large language models while training only ∼4% of the model’s parameters and reducing memory use by roughly 75%, enabling development on low-cost consumer-grade hardware.- 17 June 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (12) , 6158
Open AccessMed-LLaMA3: Advancing Medical Question-Answering Through Parameter-Efficient Fine-Tuning of Large Language Models
Despite recent advances, medical question answering systems still struggle with domain-specific reasoning and data efficiency. This paper presents Med-LLaMA3, a family of medical large language models developed by parameter-efficient fine-tuning of the LLaMA-3.1 (8 billion) and LLaMA-3.2 (1 and 3 billion) architectures using quantized low-rank adaptation (QLoRA) and low-rank adaptation (LoRA) with 4-bit quantization. Beyond model training, this work contributes the following: (1) a formalized dataset curation taxonomy (source type × clinical granularity × task format) with a source-category ablation confirming that the multi-source combination drives benchmark gains beyond any single category; (2) a systematic characterization of low-rank-adaptation rank-scaling behavior for the LLaMA-3 family in the medical domain (monotonic improvement up to rank 128, with no observed plateau); and (3) statistically validated comparisons using McNemar’s test and 95% bootstrap confidence intervals. We curated a medical instruction dataset of over 1.5 million samples spanning medical examinations, clinical dialogues, and biomedical literature. Our approach trains only ∼4% of the base model’s parameters and, consistent with prior studies of parameter-efficient methods in the medical domain, achieves performance comparable to full fine-tuning at a fraction of the memory footprint. Evaluated with five in-context examples per prompt, the 8-billion-parameter model attains a mean accuracy of 75.71% across the eight medical-domain subsets of the Massive Multitask Language Understanding benchmark; improvements over the unmodified LLaMA-3.1-8B-Instruct baseline are statistically significant on the medical multiple-choice benchmark MedMCQA and, after Bonferroni correction across the eight subsets, on three subsets (Clinical Knowledge, Medical Genetics, and Nutrition), with two further subsets being significant only before correction. A structured named-entity-recognition evaluation on 100 hospital discharge summaries (macro-averaged F1 0.94; dual-annotator agreement ) provides complementary evidence of clinical-text utility. A safety mitigation pilot shows that context-disambiguation preprocessing reduces the highest-severity abbreviation-ambiguity error rate from 30% to 10% on a 30-case held-out set. These results show that parameter-efficient fine-tuning can deliver high-performance medical large language models while training only ∼4% of the model’s parameters and reducing memory use by roughly 75%, enabling development on low-cost consumer-grade hardware. - Top Cited PapersOpen Access
- 16 October 2025
- Published by Springer Nature in Neural Computing and Applications
- p. 1-77
A survey on retrieval-augmentation generation (RAG) models for healthcare applications
- 16 October 2025
- Published by Springer Nature in Neural Computing and Applications
- p. 1-77
Top Cited PapersOpen AccessA survey on retrieval-augmentation generation (RAG) models for healthcare applications
- Open Access
- 25 September 2025
- Published by Egyptian Knowledge Bank in International Journal of Intelligent Computing and Information Sciences
- Vol. 25 (3) , 107-122
A MULTI-FEATURE ACCURATE DETECTION (MFAD) APPROACH FOR LARGE LANGUAGE MODEL-GENERATEDTEXT
Advanced Large Language Models (LLMs) generate highly complex text that closely resembles human writing. However, their rapid development raises significant concerns, such as misinformation and academic cheating. As the responsible use of LLMs becomes increasingly critical, the ability to detect LLM-generated content has emerged as a critical challenge. Existing detection methods often rely on single-feature analysis, traditional feature extraction techniques, and conventional classification models. Many also require full access to the underlying models and are sensitive to variations in text length, limiting their overall effectiveness. This paper proposes a novel Multi-Feature Accurate Detection (MFAD) approach for identifying LLM-generated text by integrating syntactic and statistical attributes with high-level semantic representations. A case study using the Human ChatGPT Comparison Corpus (HC3) is conducted to evaluate the proposed architecture. MFAD comprises six phases: text preprocessing, syntactic and statistical feature extraction, text representation, semantic feature extraction, feature concatenation, and text classification. Results show that MFAD effectively distinguishes between human-written and LLM-generated text, achieving a peak confidence score of 98%, highlighting its reliability and strong performance.
- 25 September 2025
- Published by Egyptian Knowledge Bank in International Journal of Intelligent Computing and Information Sciences
- Vol. 25 (3) , 107-122
Open AccessA MULTI-FEATURE ACCURATE DETECTION (MFAD) APPROACH FOR LARGE LANGUAGE MODEL-GENERATEDTEXT
Advanced Large Language Models (LLMs) generate highly complex text that closely resembles human writing. However, their rapid development raises significant concerns, such as misinformation and academic cheating. As the responsible use of LLMs becomes increasingly critical, the ability to detect LLM-generated content has emerged as a critical challenge. Existing detection methods often rely on single-feature analysis, traditional feature extraction techniques, and conventional classification models. Many also require full access to the underlying models and are sensitive to variations in text length, limiting their overall effectiveness. This paper proposes a novel Multi-Feature Accurate Detection (MFAD) approach for identifying LLM-generated text by integrating syntactic and statistical attributes with high-level semantic representations. A case study using the Human ChatGPT Comparison Corpus (HC3) is conducted to evaluate the proposed architecture. MFAD comprises six phases: text preprocessing, syntactic and statistical feature extraction, text representation, semantic feature extraction, feature concatenation, and text classification. Results show that MFAD effectively distinguishes between human-written and LLM-generated text, achieving a peak confidence score of 98%, highlighting its reliability and strong performance.
- Open Access
- 1 June 2024
- Published by Egyptian Knowledge Bank in International Journal of Intelligent Computing and Information Sciences
- Vol. 24 (2) , 41-57
Image Steganography: A Comparative and Practical Study
Data security and privacy concerns arise from sharing sensitive data online. Methods such as digital watermarking, cryptography, and steganography are employed to protect data. Steganography excels beyond alternative methods in safeguarding data against potential threats, with superior effectiveness and discretion. It is a promising method for safely transmitting private data across an insecure channel that conceals information from viewing. The study of hiding secret information inside an image using several methods is known as image steganography. One of the main problems with image steganography techniques is their imperceptibility and large embedding capacity. In this study, a comparative practical performance analysis of three image steganography techniques (k-LSB, CAIS, and HiNet) is performed on three different datasets (DIV2K, ImageNet, and COCO) and evaluated using four different performance metrics: SSIM, PSNR, RMSE, and MAE. The experimental results revealed that HiNet consistently achieved the best performance across all metrics and datasets. On the DIV2K dataset, HiNet achieves a PSNR of 46.57 dB and an SSIM of 0.993, significantly outperforming 4bit-LSB and CAIS. Similarly, on the ImageNet and COCO datasets, HiNet demonstrates superior performance with PSNR values of 36.63 dB and 36.55 dB, and SSIM values of 0.960 and 0.961, respectively. These results indicate that HiNet, an invertible neural-network-based method, provides substantial improvements in both the concealing and revealing of secret images, making it a highly effective solution for image steganography.
- 1 June 2024
- Published by Egyptian Knowledge Bank in International Journal of Intelligent Computing and Information Sciences
- Vol. 24 (2) , 41-57
Open AccessImage Steganography: A Comparative and Practical Study
Data security and privacy concerns arise from sharing sensitive data online. Methods such as digital watermarking, cryptography, and steganography are employed to protect data. Steganography excels beyond alternative methods in safeguarding data against potential threats, with superior effectiveness and discretion. It is a promising method for safely transmitting private data across an insecure channel that conceals information from viewing. The study of hiding secret information inside an image using several methods is known as image steganography. One of the main problems with image steganography techniques is their imperceptibility and large embedding capacity. In this study, a comparative practical performance analysis of three image steganography techniques (k-LSB, CAIS, and HiNet) is performed on three different datasets (DIV2K, ImageNet, and COCO) and evaluated using four different performance metrics: SSIM, PSNR, RMSE, and MAE. The experimental results revealed that HiNet consistently achieved the best performance across all metrics and datasets. On the DIV2K dataset, HiNet achieves a PSNR of 46.57 dB and an SSIM of 0.993, significantly outperforming 4bit-LSB and CAIS. Similarly, on the ImageNet and COCO datasets, HiNet demonstrates superior performance with PSNR values of 36.63 dB and 36.55 dB, and SSIM values of 0.960 and 0.961, respectively. These results indicate that HiNet, an invertible neural-network-based method, provides substantial improvements in both the concealing and revealing of secret images, making it a highly effective solution for image steganography.
- Open Access
- 1 December 2023
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 13 (6) , 6984-6991
Approach for Enneagram personality detection for Twitter text: a case study
Understanding people’s emotions and orientations attracts researchers nowadays. Current personality detection research concentrates on models such as the big five model, the three-factor model. The Enneagram is deeper than these models for providing a comprehensive view. This theory is a unique personality model because it illustrates what drives human behavior. This recognition helps in building smarter recommendation systems and intelligent educational systems. Enneagram personalities are realized through a long questionnaire-based test. People are not concerned about doing a test because it is time-consuming. A proposed case study employs Twitter’s text to detect Enneagram personality because it requires no time or effort. The proposed case study is based on an approach that uses a combination of ontology, lexicon, and statistical technique. This proposed case study uses the biography description text and 40 tweets of a Twitter profile text. The highest probability percentage is peacemaker personality which is 15.58%. This result means that the identified personality is the peacemaker. The outcome is equivalent to the determination of the Enneagram’s specialized people. This result promises more positive outcomes. This is the first automated approach to determine the Enneagram from text.
- 1 December 2023
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 13 (6) , 6984-6991
Open AccessApproach for Enneagram personality detection for Twitter text: a case study
Understanding people’s emotions and orientations attracts researchers nowadays. Current personality detection research concentrates on models such as the big five model, the three-factor model. The Enneagram is deeper than these models for providing a comprehensive view. This theory is a unique personality model because it illustrates what drives human behavior. This recognition helps in building smarter recommendation systems and intelligent educational systems. Enneagram personalities are realized through a long questionnaire-based test. People are not concerned about doing a test because it is time-consuming. A proposed case study employs Twitter’s text to detect Enneagram personality because it requires no time or effort. The proposed case study is based on an approach that uses a combination of ontology, lexicon, and statistical technique. This proposed case study uses the biography description text and 40 tweets of a Twitter profile text. The highest probability percentage is peacemaker personality which is 15.58%. This result means that the identified personality is the peacemaker. The outcome is equivalent to the determination of the Enneagram’s specialized people. This result promises more positive outcomes. This is the first automated approach to determine the Enneagram from text.
- 18 March 2023
- Published by Springer Nature
- p. 1-9
Enneaontology: Toward an Enneagram Personality Detection
- 18 March 2023
- Published by Springer Nature
- p. 1-9
Enneaontology: Toward an Enneagram Personality Detection
- Open Access
- 26 February 2023
- Published by Egyptian Knowledge Bank in International Journal of Intelligent Computing and Information Sciences
- p. 1-11
ARCHITECTURE FOR PERSONALITY DETECTION USING ENNEAGRAM KNOWLEDGE: CASE STUDY
Researchers are concerned with automated personality detection from social media. Automated Personality detection from text benefits in social media like: attracting more users, career advising and getting more advertisements. Traditional personality detection is done by using an assessment test. Performing a test is time-consuming so users aren’t interested in taking a test. This paper presents case study and architecture on automated personality from text using twitter text. The case study uses public text to identify the personality of the profile. The applied personality Model is the Enneagram model. Proposed architecture contains four phases: text preprocessing, feature extraction, feature selection and personality detection. Feature selection is done by using Enneagram ontology and lexicon. Personality detection is utilized by using statistical approaches. The Enneagram knowledge is modeled using ontology. The lexicon is a source to enrich the ontology seed. Statistical Approach is utilized to identify the personality. The case study identifies the personality. The highest outcome percentage is “investigator” personality, which is 24 %. This indicates that the personality is an investigator. This result is similar to official Enneagram experts’ analysis. This model is the first one which uses the Enneagram model as automatic detection. Enneagram is a powerful personality model that aids psychiatrists and physicians to understand the patient's personality intensely. This knowledge gives them the tools to support and aid the patient to heal faster. The promising outcomes open the door to further research in this area.
- 26 February 2023
- Published by Egyptian Knowledge Bank in International Journal of Intelligent Computing and Information Sciences
- p. 1-11
Open AccessARCHITECTURE FOR PERSONALITY DETECTION USING ENNEAGRAM KNOWLEDGE: CASE STUDY
Researchers are concerned with automated personality detection from social media. Automated Personality detection from text benefits in social media like: attracting more users, career advising and getting more advertisements. Traditional personality detection is done by using an assessment test. Performing a test is time-consuming so users aren’t interested in taking a test. This paper presents case study and architecture on automated personality from text using twitter text. The case study uses public text to identify the personality of the profile. The applied personality Model is the Enneagram model. Proposed architecture contains four phases: text preprocessing, feature extraction, feature selection and personality detection. Feature selection is done by using Enneagram ontology and lexicon. Personality detection is utilized by using statistical approaches. The Enneagram knowledge is modeled using ontology. The lexicon is a source to enrich the ontology seed. Statistical Approach is utilized to identify the personality. The case study identifies the personality. The highest outcome percentage is “investigator” personality, which is 24 %. This indicates that the personality is an investigator. This result is similar to official Enneagram experts’ analysis. This model is the first one which uses the Enneagram model as automatic detection. Enneagram is a powerful personality model that aids psychiatrists and physicians to understand the patient's personality intensely. This knowledge gives them the tools to support and aid the patient to heal faster. The promising outcomes open the door to further research in this area.