Predictive Modeling of Cancer Progression

There are two principal challenges of applying ML in cancer progression modeling: Having enough phenotypically rich data to train models and professional validation of generated insights. So we had all the ingredients to deal with the scientific problem and achieve the objective.

• In the majority of cases, medical data is highly disorganized. Thus the first step was to convert scattered medical definitions to strict machine-readable language using extensive domain expertise in healthcare, SQL, and NLP tools (Python-based).
• Then we implemented deep machine learning technologies (CNN-based models) to make predictions.
• And finally, we visualized obtained models using R and Shiny application in a human-readable format.

1. Domain Expertise in Healthcare
2. Natural Language Processing (NLP)
3. Relational Database Management (SQL-based)
4. Data Science and Data Analytics
5. Programming languages: R, Python, SQL
6. Machine Learning
7. Predictive Analysis

We created a predictive model to forecast the risk of a disease progression (e.g., relapse, protracted clinical course) for patients with lung cancer and lymphoma: After transforming raw data to the OMOP CDM data standard, we defined the required cohorts. Based on it, we applied CNN-based deep learning models, predicting cancer outcomes to data collected for patients during the "time at risk" window.

As a result, we contributed to the early diagnosis of lung cancer and lymphoma with an ML-driven cancer progression predictive model.