Inspiration
My previous work as an Oncology Medical Assistant has given me the experience of understanding the daily workflow of an oncologist firsthand. We often ordered Complete Blood Count (CBC) tests for patients which are critical in monitoring treatment side effects and the potential spread of cancer. While Oncologists are highly trained in reading the values (ex. WBC, ANC, RBC, HGB, PLT, LYM, MONO) in these tests, it is easy to become overwhelmed due to needing to monitor hundreds of patients with multiple CBC tests, with patients having different baseline values and risk levels. This tool was created to help oncologists organize and track the CBC values for patients diagnosed with cancer over a comprehensive time period.
Function
This tool uses a multi-input Temporal Convolutional Network trained on synthetic patient data grounded in published clinical reference ranges and chemotherapy pharmacology. By predicting the severity of anomaly risk in each patient in relation to specific CBC variable values, this tools allows Oncologists to make informed decisions on whether or not to switch or keep current cancer treatment plans.
Relevant Skills
PyTorch, scikit-learn, SHAP, TCN, Streamlit, Plotly