This repository provides the official implementation of the framework proposed in the IEEE paper: "A Clinical Informatics Framework for Myeloid Oncology: Scalable AI and LLM Integration for Adaptive Trial Management and Automated FDA Compliance".
The advancement of treatment strategies in myeloid malignancies (such as Acute Myeloid Leukemia - AML) relies heavily on precision oncology, dynamic biomarker profiling, and adaptive trial designs. This project introduces an Artificial Intelligence-enabled clinical trial framework designed to navigate the complexities of molecularly defined eligibility criteria, predictive safety monitoring, and the rigorous demands of accelerated FDA regulatory submissions.
Precision Oncology, Myeloid Malignancies, Acute Myeloid Leukemia (AML), Adaptive Clinical Trials, Machine Learning in Healthcare, Electronic Data Capture (EDC), Regulatory Automation, Large Language Models (LLMs), FDA 21 CFR Part 11, MyeloMATCH.
This repository translates the conceptual clinical informatics architecture into a deployable, code-driven framework:
- Data Acquisition & Harmonization (Synthetic): Generation of high-dimensional clinical and biomarker datasets (FLT3, IDH1/2, NPM1) mimicking the Beat AML and TCGA-LAML open-source cohorts.
- Biomarker Interpretation & Protocol Rule Engine: Automated mapping of patient morphology and cytogenetics against protocol-specific thresholds.
- Patient Stratification (GBDT): Gradient Boosted Decision Trees deployed for real-time, explainable (SHAP-integrated) trial matching and eligibility assignment.
- Temporal Response Prediction (LSTM): Long Short-Term Memory networks designed to forecast Day 28 early treatment response (CR/CRi) using sequential hematological clearance kinetics.
- Adverse Event Modeling (Regularized Logistic Regression): Robust, clinically interpretable risk forecasting for Grade 3-4 hematologic toxicities (e.g., Neutropenia).
- Agentic AI & Regulatory Submission Engine: An LLM-driven automation layer that maps ML outputs to FDA Predetermined Change Control Plans (PCCP) and 21 CFR Part 11 compliance standards, outputting structured clinical recommendations.
The core implementation is housed in MyeloidOncology_AI_Framework.ipynb. The notebook is meticulously documented, mapping directly to the paper's headings:
- Clinical Data Analysis and Methods (Data Harmonization)
- Machine Learning Model Design
- Eligibility Stratification (GBDT)
- Temporal Response Prediction (LSTM)
- Adverse Event Forecasting (LogReg)
- Agentic Automation Layer (LLM & FDA Compliance)
- ML Model Performance and Benchmarking
If you utilize this framework or code in your research, please cite the original IEEE paper:
IEEE Format: S. Vijayakumar, S. P. Kane, S. Senthilkumar, P. Vaiayapuri, and F. Louis, "A Clinical Informatics Framework for Myeloid Oncology: Scalable AI and LLM Integration for Adaptive Trial Management and Automated FDA Compliance," IEEE, 2026. Available: https://ieeexplore.ieee.org/document/11378528
Authors:
- Senthilkumar Vijayakumar (IEEE Senior Member)
- Shaunak Pai Kane (IEEE Member)
- Selvavaani Senthilkumar
- Dr. Parameshwari Vaiayapuri, MBBS
- Filious Louis (IEEE Senior Member)
Install the required Python dependencies:
pip install pandas numpy scikit-learn xgboost torch shap matplotlib seaborn jupyter- Clone the repository.
- Run the synthetic data generator (if you wish to regenerate the cohort):
python3 generate_data.py - Launch the Jupyter Notebook to explore the models, SHAP explanations, and Agentic AI outputs:
jupyter notebook MyeloidOncology_AI_Framework.ipynb