Skip to content
View MahshadSa's full-sized avatar

Block or report MahshadSa

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
MahshadSa/README.md

Mahshad Sarikhani

MD and research associate in abdominal radiology at Shariati Hospital, Tehran University of Medical Sciences. I work on quantitative MRI in inflammatory bowel disease, at the point where imaging methodology meets clinical decision-making.

Most of what I build starts from a problem I hit in my own research: disease activity that endoscopy measures invasively and MRE could measure better, radiology reports that hold structured information no one can query, literature that arrives faster than I can triage it. The tools here are reproducible by design, because the alternative is work that cannot be checked.

Current focus

  • Bayesian multilevel modelling for segmental Mayo Endoscopic Score triage in ulcerative colitis, using MRE features and faecal calprotectin
  • Quantitative MRE in ulcerative colitis: motility, chronicity, and activity
  • Working toward an agentic LLM pipeline for Lémann Index scoring from free-text MRE reports in Crohn's disease, currently at the deterministic foundation stage

Recent

  • Two first-author oral presentations at ESGAR 2026, both delivered on my behalf by Dr Gauraang Bhatnagar
  • First-author motility manuscript under review at Journal of Crohn's and Colitis
  • Ongoing collaboration with UCL and Motilent on quantitative bowel MRE

Selected projects

uc-mre-activity-risk-stratifier — Bayesian ordinal regression estimating segment-level Mayo endoscopic severity from MRE features and faecal calprotectin. Marginalises the patient random intercept via Gauss-Hermite quadrature to give new-patient probabilities rather than population averages. Across 245 segments from 49 patients, marginal AUC 0.84 to 0.87; 53% of segments resolve into definitive low or high-risk tiers without endoscopy. Model constants are versioned; scope limits are explicit. DOI: 10.5281/zenodo.18171572

abdominal-radiology-reports — Pipeline for turning free-text abdominal radiology reports into structured, queryable records, built as the data infrastructure the imaging work runs on. Patient data never enters the repository; tests use synthetic documents only. Phase 0 complete, Phase 1 in progress.

lemann-li-app — Streamlit implementation of the updated Organ and Global Lémann Index (Pariente et al., Gastroenterology 2021). Research and education use only. DOI: 10.5281/zenodo.17513800

ibd-digest — A daily pipeline that fetches new IBD imaging research from PubMed and Crossref, ranks it by SPECTER2 semantic similarity against a curated corpus, and delivers a tiered digest to Obsidian. The interesting problem was calibration: SPECTER2 produces uniformly high cosine similarities on in-domain biomedical text, so fixed thresholds are useless. Tiers are anchored to corpus percentiles instead. Runs unattended on a schedule.

radiology-clip-mini — A CLIP-style image-text retrieval baseline for chest radiographs and reports, built as a reference implementation others can actually run. ResNet-18 and DistilBERT encoders, symmetric InfoNCE, Grad-CAM interpretability, patient-level splits to prevent leakage. Versioned on Zenodo with CI and deterministic seeds. DOI: 10.5281/zenodo.17795012

Google Scholar | ORCID

Pinned Loading

  1. uc-mre-activity-risk-stratifier uc-mre-activity-risk-stratifier Public

    Segment-level UC activity risk stratification from MRE using Bayesian ordinal modeling

    Python

  2. abdominal-radiology-reports abdominal-radiology-reports Public

    Reproducible pipeline for structuring and de-identifying abdominal radiology reports for clinical NLP research

    Python

  3. radiology-clip-mini radiology-clip-mini Public

    Compact CLIP-style baseline for chest X-ray and report retrieval, built as a reproducible reference implementation

    Python

  4. lemann-li-app lemann-li-app Public

    Streamlit web-app for calculating the updated Lémann Index, measures cumulative bowel-damage in Crohn’s disease (strictures + fistulae + surgery). Research/education use only.

    Python

  5. ibd-digest ibd-digest Public

    Daily digest of new IBD imaging research, ranked by SPECTER2 semantic similarity and delivered to Obsidian.

    Python