End-to-end clinical-NLP replication of three published methods on a real 30-patient synthetic oncology cohort. Trajectory smoothing, performance-status extraction, and inflection-point detection across radiology and ECOG signals — every extraction citing a verbatim sentence from source text.
Live API ↗No single paper does all three on one patient. ONCOTRAJ stitches them into one longitudinal pipeline.
Kernel-smoothed time-series of radiologic scoring across 18-month windows. Per-report extraction of cancer_present / progression / response / metastasis from chest CT, MRI, PET-CT reports, with a verbatim citing sentence drawn from the source.
Kehl K.L. et al. — Nature Communications, 2024 · DOI 10.1038/s41467-024-54071-xECOG and KPS extraction from medical-oncology notes via four prompting strategies — regex, simple prompting, Chain-of-Thought, and Double Filtering — reproducing the comparison table from the original paper on the same cohort.
Kehl group — JCO CCI, Feb 2026 · DOI 10.1200/CCI-25-00226Change-point detection on smoothed trajectories. Fires when radiology shows progression and ECOG worsens to ≥ 2 within a 60-day window, with a 90-day refractory between events — flagging candidates for treatment change or trial referral.
Kehl K.L., Schrag D. et al. — JCO CCI, 2020 · DOI 10.1200/CCI.20.00184Pick a patient. Radiology score (crimson, left axis) and ECOG performance (phosphor, right axis) over 18 months. Dashed line marks the detected inflection point.
@misc{sami2026oncotraj,
author = {Sami, Abdullah Abdul},
title = {ONCOTRAJ: Oncology Trajectory & Inflection Toolkit},
year = {2026},
url = {https://oncotraj-toolkit.pages.dev}
}