Progress 08/15/24 to 08/14/25
Outputs
Target Audience:During the reporting period untilAugust 31, 2025, our project efforts focused on engaging and informing a broad spectrum of target audiences within the fields of food safety, artificial intelligence (AI), and computational biology. These audiences included academic researchers, students, industry professionals, and interdisciplinary scientists interested in the application of AI for food safety. We reached researchers and scholars in fields such as food safety, microbiology, AI, computing vision, and image analysis through active dissemination of our work. Our team will presentresearch findings at major academic conferences, including the ICPP Companion '25: 54th International Conference on Parallel Processing Companion, San Diego, CA, USA, 8-11 September 2025 at the HARVEST Workshop, where our paper "XAIPath: Explainable AI for Salmonella Detection in Microscopic Onion Images" was accepted and well-received. Additionally, we presented as a position paper "From Pixel to Policy: AI-Driven Microbial Detection" at the BRICCs Research Data Management Conference, allowing us to engage in knowledge exchange and obtain valuable feedback from multidisciplinary researchers, thus contributing to scientific progress and encouraging collaboration. In addition, we submitted"Background-Aware Instance Segmentation for Early Detection of E. coli and Salmonella in Time-Stamped Microscopy Images", 24th International Conference on Machine Learning and Applications (AMLA), (Submitted) Dec 3-4, 2025. In addition, our participation in the ByteBoost Workshop at the University of Pittsburgh and the BIRCC and PACES Workshops 2025 with Texas A&M University--College Station, TX,enabled us to engage with technical audiences focused on scalable AI solutions. We presented our BactoDetect system, a high-performance bacterial detection platform using YOLOv8, targeting rapid and real-time pathogen identification in microscopy images. These presentations reached researchers and developers from the HPC and AI optimization sectors, particularly those working with NVIDIA H100 GPUs (on ACES HPC) and ARM-based A64FX (Ookmi) systems. During this period, a graduate student (June 2 to August 2, 2025) and a post-doctoral researcher (October 15, 2024, to August 31, 2025) were actively involved in data annotation, image acquisition, and early-stage modeling. They received hands-on training through research-based learning and exposure to interdisciplinary computational approaches. Media Coverage This project received national media coverage and public engagement: Food Safety Magazine USDA Awards Researcher $150,000 to Develop Rapid AI Detection System for Salmonella on Onions Food Safety News SIU Researcher Gets $150K Grant to Use AI for Salmonella Detection in Onions SIU Newsroom Feature Article SIU Researcher Gets $150K to Use AI to Prevent Food Poisoning Television Interview TV segment aired locally (12KFVS), where PI Dr. Anas Alsobeh discussed the project in the context of a recent McDonald's onion-linked E. coli outbreak. The interview emphasized the technology's potential to prevent similar events via rapid, AI-powered screening. Link: https://www.kfvs12.com/2024/10/30/siu-researchers-studying-how-ai-can-prevent-food-poisoning/ Invited Talks & Outreach Invited Speaker: Utah Onion Association Annual Meeting (2026) Host: Dr. Milena Oliveira, Utah State University Extension Topic: AI for Real-Time Pathogen Detection in Onion Processing Facilities Goal: Share results with growers and food safety specialists Date: Feb 2026. Planning underway for thefirst workshopto occur after full image annotation and preliminary results are analyzed. Target attendees include food safety specialists, agricultural researchers, and processors. In preparation for our first educational workshop (scheduled post-data collection), we began internal planning involving students and research assistants. Although the workshop has not yet been delivered during this reporting period, planning efforts have focused on curriculum design, outreach materials, and platform selection to maximize impact across stakeholder groups, including extension educators and food industry personnel. Overall, our engagement strategies during this period successfully reached stakeholders in academia, high-performance computing, and microbiology/food safety, laying a strong foundation for future outreach and technology transfer efforts aimed at translating this research into real-world applications. Changes/Problems:1. Data Collection Expansion Beyond Initial Scope (Positive Deviation) The number of microscopic images captured exceeded the original plan (from 1,800 to over 3,800). The initial imaging process yielded more high-quality, time-stamped samples than anticipated, especially across additional onion conditions and incubation times.This change enhanced the robustness of the dataset, increasing the training potential for AI models. No negative impact on budget or timeline; the effort was absorbed through efficient resource use and scheduling. 2. Extended Annotation Timeline Manual image annotation using CVAT took longer than originally anticipated due to the complexity of differentiating overlapping microcolonies and accurately tracking growth stages. High-resolution images with varying visual conditions required greater precision in bounding and polygon annotations for model training. This slightly delayed the handoff of the dataset to the modeling team, pushing initial deep learning experimentation by approximately 3-4 weeks. However, the impact was mitigated through parallel progress on model architecture design. 3. Change in Model Selection Strategy Initially, the project planned to use YOLOv4 for object detection. However, after preliminary experimentation, the team adopted YOLOv8 and added support for segmentation models like Mask R-CNN and CellPose. YOLOv8 provided improved speed, flexibility, and modularity. Mask R-CNN and CellPose were added to address detection of small and overlapping microcolonies, which YOLO alone could not adequately separate.This change has improved model performance and aligns better with the complexity of the biological data. Computational resource usage increased, but the university's HPC system accommodated this shift without additional cost. 4. Workshop Rescheduling The Year 1 educational workshop originally planned for late spring 2025 has been rescheduled to fall 2025. This adjustment was made to align with university scheduling, ensure participant availability, and allow incorporation of more finalized AI models into training content. No long-term effects on project outcomes. The delay will enhance the effectiveness of the workshop by allowing demonstration of more advanced tools. 5. Hire Graduate Student Assignment The graduate student was assigned to the project before June 2025. This isnot an effect of the project budget. With both researchers now active, the team is operating at full capacity. What opportunities for training and professional development has the project provided?The project has provided multiple training and development opportunities: A full-time post-doctoral researcher joined the project in October 2024 and has since received comprehensive interdisciplinary training that includes: Gained experience in purchasing lab equipment (stomacher, microscope). Hands-on experience in preparing and imaging contaminated onion samples using phase contrast and fluorescence microscopy techniques. Learning to annotate time-stamped microscopic images for AI training, organize large image datasets, and follow FAIR data practices. Participated in the design, implementation, and testing of real-time AI models, including YOLOv8 and segmentation models for detecting Salmonella and E. coli. Contributed to the preparation and submission of two academic papers, enhancing public speaking and scholarly writing skills. Participated in collaborative data processing and mentored a graduate student. A graduate student joined in June 2025 and engaged in structured learning experiences related to: Leadefforts in dataset annotation using CVAT software Introduction to object detection techniques and explainable AI (XAI) frameworks for foodborne pathogen analysis. Training in handling biological samples and operating imaging systems. Exposure to tools such as Python, LabelImg, and deep learning libraries (e.g., PyTorch, TensorFlow). Assisted in preparing conference presentations and workshop content for upcoming public outreach and stakeholder training events. Assisted in the HARVEST workshopand was awarded $1,500 to attend and present a poster. The team collaboratively prepared for the first hands-on training workshop on AI in food safety. This process offered learning opportunities in curriculum design, stakeholder communication, and translational research presentation. How have the results been disseminated to communities of interest?Conference Presentations and Submissions: "XAIPath: Temporal-Environmental Explainable AI Framework for Co-Contaminated Food Pathogen Detection in Microscopic Imaging" was accepted and will be presented at the 54th International Conference on Parallel Processing Companion (ICPP Companion '25) in September 2025. "From Pixels to Policy: Explainable AI for Microbial Detection in Food" was presented at the BRICCs Conference 2025 to an audience of many different researchers and data scientists. "Background-Aware Instance Segmentation for Early Detection of E. coli and Salmonella in Time-Stamped Microscopy Images" was submitted to the 24th International Conference on Machine Learning and Applications (ICMLA) for December 2025. Project progress and impact received significant public exposure through media outlets and institutional platforms: Featured in Food Safety Magazine (January 2025): "USDA Awards Researcher $150,000 to Develop Rapid AI Detection System for Salmonella on Onions." https://www.food-safety.com/articles/10084-usda-awards-researcher-150-000-to-develop-rapid-ai-detection-system-for-salmonella-on-onions Covered by Food Safety News: "SIU Researcher Gets $150,000 Grant to Use AI for Salmonella Detection in Onions."https://www.foodsafetynews.com/2025/01/siu-researcher-gets-150000-grant-to-use-ai-for-salmonella-detection-in-onions/ Highlighted by Southern Illinois University (SIU) in an official press release:https://news.siu.edu/2024/10/102924-siu-researcher-gets-150k-to-use-ai-to-prevent-food-poisoning.php A TV interview and university e-newsletters further amplified project visibility. Stakeholder Engagement PI Anas Alsobeh was invited to speak at the Utah Onion Association Meeting (2026) by Extension Vegetable Specialist Dr. Milena Oliveira (Utah State University), opening a potential collaboration with onion growers in Utah. What do you plan to do during the next reporting period to accomplish the goals?We plan to: Develop and Optimize AI Models for Bacteria Detection Finalize pre-processing of annotated image data (using CVAT Box and Pylogon tools). Train and evaluate segmentation and detection models (YOLOv8, ResNet, Mask R-CNN, ViT) on Salmonella and E. coli images. Perform model optimization to increase detection accuracy and reduce false positives/negatives, using cross-validation, hyperparameter tuning, and ensemble learning. Integrate time-stamped metadata (growth stage, temperature, incubation conditions) for temporal-aware detection models. Leverage institutional HPC clusters for efficient training and integrate XAI tools like Grad-CAM for model interpretability. Explore modular deployment options for food industry use (cloud + edge AI). Expand Stakeholder Engagement and Outreach Prepare for invited talk at the Utah Onion Association Annual Meeting, engaging growers and processors interested in emerging food safety technologies. Develop outreach materials including explainer videos and web content to simplify understanding of the AI detection system for non-technical stakeholders. Organize and Conduct Educational Workshops Host a hands-on AI and Imaging Workshop at Southern Illinois University for students, faculty, and food safety professionals, introducing them to: Microscopy image processing. AI model design for bacterial detection. Best practices for data annotation and model evaluation. Offer training modules for food scientists and lab technicians focused on how to integrate AI into routine food inspection processes. Assess participant knowledge and feedback to refine future training. Continue the dissemination of results. Present accepted papers and ongoing research at two major conferences: ICPP Companion 2025 (XAIPath paper, September 2025). ICMLA 2025 (Pending acceptance for segmentation model paper, December 2025). Submit an additional manuscript to a peer-reviewed journal focused on AI in food safety or applied microbiology. Publish dataset documentation and codebaseto Ag Data Commons and GitHub with digital persistent identifiers (DOIs) and proper attribution to USDA-NIFA funding. Advance Outreach and Partnerships: Maintain communication with Utah State University Extension for collaboration.
Impacts
What was accomplished under these goals?
1. Dataset Development and Image Annotation Successfully collected and cultured Salmonella Thompson and E. coli K-12 under varying conditions to simulate realistic contamination environments.We acquired over 3,800 high-resolution microscopic images using the Olympus IX70 Inverted Microscope, exceeding the original target of 1,800 images. These are images:(1)Salmonellawithoutonion (2)Salmonellawith onion (3) E. coli without onion (4) E. coli with onion. These collected over a growth period of 0.5 to 4 hours. We annotated bacterial microcolonies using CVAT (Computer Vision Annotation Tool), categorizing them by strain, onion presence, and growth stage. The dataset supports both classification and instance segmentation tasks using deep learning models. 2. Development of AI Models for Pathogen Detection A prototype object detection model using YOLOv8 was trained on the annotated dataset. Initial results indicate promising accuracy (>92%) in identifying early-stage microcolonies. A prototype object detection model was developed using YOLOv8. Preliminary results show >92% accuracy in identifying early-stage microcolonies. To improve model robustness, we applied background-aware segmentation and augmentation techniques to address image noise and overlapping colonies. Submitted: "Background-Aware Instance Segmentation for Early Detection of E. coli and Salmonella in Time-Stamped Microscopy Images" to the 24th International Conference on Machine Learning and Applications (ICMLA). Accepted: "XAIPath: Temporal-Environmental Explainable AI Framework for Co-Contaminated Food Pathogen Detection in Microscopic Imaging," to be presented at the ICPP Companion 2025. In Press: "From Pixels to Policy: RDM Strategies for AI-Driven Bacterial Detection in Food Safety Research," accepted to the BRICCs-RDM Conference in Alexandria, VA (July 2025) 3. Dissemination and Outreach Our work was presented at multiple national conferences and workshops: BRICCs Conference 2025: "From Pixel to Policy" ICPP Companion 2025: "XAIPath" (accepted) ByteBoost HPC Workshop: "BactoDetect: High-Performance Object Detection of Bacteria in Microscopic Images Using AI" Public and media engagement included: Coverage by Food Safety Magazine and Food Safety News University-wide press release from SIU Local TV coverage Formal invitation to present at the 2026 Utah Onion Association Annual Meeting 4.Education and Workforce Development Trained a post-doctoral researcher and the graduate students on advanced imaging and AI-based bacterial detection methods. Prepared workshop materials to be delivered in the next phase, integrating AI in food inspection processes. The graduate student and post-doc gained hands-on experience in microscopy imaging, annotation, and model development (YOLO, ViT, CellPose), preparing them for independent research roles.
Publications
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2025
Citation:
AlSobeh, A., AbuGhazaleh, A., Dhahir, N., Rababa, M. "XAIPath: Temporal-Environmental Explainable AI Framework for Co-Contaminated Food Pathogen Detection in Microscopic Imaging." 54th International Conference on Parallel Processing Companion (ICPP '25 Companion), San Diego, CA, September 811, 2025.
Status: Accepted.
https://camps.aptaracorp.com/ACM_PMS/PMS/ACM/ICPPCOMPANION25/30/5b06d196-72fc-11f0-957d-16ffd757ba29/OUT/icppcompanion25-30.html#
- Type:
Conference Papers and Presentations
Status:
Submitted
Year Published:
2025
Citation:
Koirala, B., AlSobeh, A., Dhahir, N., AbuGhazaleh, A. "Background-Aware Instance Segmentation for Early Detection of E. coli and Salmonella in Time-Stamped Microscopy Images." Submitted to: 24th International Conference on Machine Learning and Applications (ICMLA), December 34, 2025.
Status: Submitted.
- Type:
Other
Status:
Awaiting Publication
Year Published:
2025
Citation:
Anas AlSobeh, Malek Rababah, Namriq Dhar, Amer Abu Gazaleh, 2025. From Pixels to Policy: RDM Strategies for AI-Driven Bacterial Detection in Food Safety Research. Research Data Management (BRICCs-RDM) Conference, the Westin Alexandria Old Town, 400 Courthouse Square, Alexandria, VA 22314. July 9-11, 2025. (On press).
|