Engineering Explainable AI Systems for Clinical Impact

Medical AI Research • Transfer Learning • Computer Vision

ICU treatment timing system trained on MIMIC-III data. A2C agent achieved 99.5% clinical appropriateness with +76–145 reward improvement over baselines. Implemented rule-based safety filter boosting PPO/DQN appropriateness by 40 points. Custom 26D Gym environment with temporal lab trends and treatment effect simulation.

Stable-Baselines3GymNumPyPandasMIMIC-III

Clinical-grade diagnostic system achieving 99% four-class tumor classification with radiologist-validated interface. End-to-end DICOM pipeline development investigating optimal preprocessing strategies for limited medical imaging datasets. Deployment-ready architecture with clinical workflow integration.

EfficientNet-B7PyDICOMPyQt5SQL

Automated hematology platform for real-time RBC/WBC/platelet detection and classification of leukemia, malaria, and anemia. Reduced diagnostic time from 45 minutes to 3 minutes. Pipeline validated by clinical hematologist. Featured on BBC News Arabic — 4Tech program.

YOLOv8U-NetOpenCVPyTorch

Conv1D neural network achieving R² = 0.88 for insurance cost forecasting. Feature engineering with SHAP analysis identified key cost drivers. Systematic ablation study investigating optimal temporal convolution strategies for healthcare prediction tasks.

Conv1DSHAPScikit-learnPlotly

Presented transfer learning optimization methodologies for medical image classification in resource-limited clinical environments. Investigated disease-specific architecture selection achieving 90–99% accuracy across five clinical domains. Awarded Outstanding Research Presentation.

★ Best Presentation AwardTransfer LearningDeep LearningMedical Imaging

Multi-disease diagnostic platform with 90-99% accuracy across five disease domains. Systematic comparative evaluation of 12 deep learning architectures using stratified cross-validation. Transfer learning investigation for multi-domain medical classification.

TensorFlowResNetEfficientNetFlask

About

Developing clinically-deployable AI systems that bridge academic research and healthcare impact. My work investigates explainable deep learning architectures, transfer learning optimization, and diagnostic system design for resource-constrained clinical environments.

Start a Conversation

Contact me

LINKEDIN GITHUB EMAIL

Developed by Ahmed Messaad

Engineering Explainable AI Systems for Clinical Impact

Medical AI Research • Transfer Learning • Computer Vision

About

Developing clinically-deployable AI systems that bridge academic research and healthcare impact. My work investigates explainable deep learning architectures, transfer learning optimization, and diagnostic system design for resource-constrained clinical environments.

Start a Conversation

Contact me

LINKEDIN GITHUB EMAIL

Developed by Ahmed Messaad

ICU treatment timing system trained on MIMIC-III data. A2C agent achieved 99.5% clinical appropriateness with +76–145 reward improvement over baselines. Implemented rule-based safety filter boosting PPO/DQN appropriateness by 40 points. Custom 26D Gym environment with temporal lab trends and treatment effect simulation.

Stable-Baselines3GymNumPyPandasMIMIC-III

Clinical-grade diagnostic system achieving 99% four-class tumor classification with radiologist-validated interface. End-to-end DICOM pipeline development investigating optimal preprocessing strategies for limited medical imaging datasets. Deployment-ready architecture with clinical workflow integration.

EfficientNet-B7PyDICOMPyQt5SQL

Automated hematology platform for real-time RBC/WBC/platelet detection and classification of leukemia, malaria, and anemia. Reduced diagnostic time from 45 minutes to 3 minutes. Pipeline validated by clinical hematologist. Featured on BBC News Arabic — 4Tech program.

YOLOv8U-NetOpenCVPyTorch

Conv1D neural network achieving R² = 0.88 for insurance cost forecasting. Feature engineering with SHAP analysis identified key cost drivers. Systematic ablation study investigating optimal temporal convolution strategies for healthcare prediction tasks.

Conv1DSHAPScikit-learnPlotly

Presented transfer learning optimization methodologies for medical image classification in resource-limited clinical environments. Investigated disease-specific architecture selection achieving 90–99% accuracy across five clinical domains. Awarded Outstanding Research Presentation.

★ Best Presentation AwardTransfer LearningDeep LearningMedical Imaging

Multi-disease diagnostic platform with 90-99% accuracy across five disease domains. Systematic comparative evaluation of 12 deep learning architectures using stratified cross-validation. Transfer learning investigation for multi-domain medical classification.

TensorFlowResNetEfficientNetFlask

Engineering Explainable AI Systems for Clinical Impact

Medical AI Research • Transfer Learning • Computer Vision

About

Developing clinically-deployable AI systems that bridge academic research and healthcare impact. My work investigates explainable deep learning architectures, transfer learning optimization, and diagnostic system design for resource-constrained clinical environments.

ICU treatment timing system trained on MIMIC-III data. A2C agent achieved 99.5% clinical appropriateness with +76–145 reward improvement over baselines. Implemented rule-based safety filter boosting PPO/DQN appropriateness by 40 points. Custom 26D Gym environment with temporal lab trends and treatment effect simulation.

Stable-Baselines3GymNumPyPandasMIMIC-III

Clinical-grade diagnostic system achieving 99% four-class tumor classification with radiologist-validated interface. End-to-end DICOM pipeline development investigating optimal preprocessing strategies for limited medical imaging datasets. Deployment-ready architecture with clinical workflow integration.

EfficientNet-B7PyDICOMPyQt5SQL

Automated hematology platform for real-time RBC/WBC/platelet detection and classification of leukemia, malaria, and anemia. Reduced diagnostic time from 45 minutes to 3 minutes. Pipeline validated by clinical hematologist. Featured on BBC News Arabic — 4Tech program.

YOLOv8U-NetOpenCVPyTorch

Conv1D neural network achieving R² = 0.88 for insurance cost forecasting. Feature engineering with SHAP analysis identified key cost drivers. Systematic ablation study investigating optimal temporal convolution strategies for healthcare prediction tasks.

Conv1DSHAPScikit-learnPlotly

Presented transfer learning optimization methodologies for medical image classification in resource-limited clinical environments. Investigated disease-specific architecture selection achieving 90–99% accuracy across five clinical domains. Awarded Outstanding Research Presentation.

★ Best Presentation AwardTransfer LearningDeep LearningMedical Imaging

Multi-disease diagnostic platform with 90-99% accuracy across five disease domains. Systematic comparative evaluation of 12 deep learning architectures using stratified cross-validation. Transfer learning investigation for multi-domain medical classification.

TensorFlowResNetEfficientNetFlask

Start a Conversation

Contact me

LINKEDIN GITHUB EMAIL

Developed by Ahmed Messaad