GSoC 2026 · OSIPI

Quality Control ToolBox
v1.0 for ASL MRI

54 automated checks across brain, kidney and placenta. Every scan gets a PASS, WARN or FAIL — with the reason, and the source behind it.

20brain
19kidney
15placenta
479tests
Agnik MisraContributor
María Mora Álvarez · Sudipto DoluiMentors

Organ 1 of 3

Brain 20 checks

qc-design.md ↗

The v1.0 target, and the only organ with a published quality index behind it.

1 · Papers

  1. 1Dolui 2024Module 1 QEI: the one metric with a published 0.5 cutoff
  2. 2Alsop 2015GM CBF 40-100 band, M0 TR 5 s rule, M0-no-BS, acquisition constants
  3. 3Adebimpe 2022GM/WM ratio > 1, mean FD > 1 mm, tissue mask 0.7, QEI code
  4. 4Power 2012Module 7 motion: framewise displacement, 0.5 mm censoring, 50 mm head radius
  5. 5Mutsaerts 2017Check 2.1 spatial CoV metric and its arterial-transit-time meaning
  6. 6Wu 2013WM CBF PASS band 15.8-27.5; the only published WM numbers
  7. 7Mutsaerts 2020The sCoV 0.67 / 1.0 verdict tiers (an implementation convention)
  8. 8Clement 2022Module 5 schema: the required-field list check 5.1 enforces

Read firstthe eight that decided the design

2 · Worked by hand

Worked by handevery formula, before any code

3 · Architecture & docs

Architecture & docsthe check map, then the document

4 · Live report

Live reportreal PASS output, not a mockup

Organ 2 of 3

Kidney 19 checks

kidney/KIDNEY_QC_DESIGN.md ↗

ASL is used in the kidney every day. No quality-control tool existed for it.

1 · Papers

  1. 1Nery 2020The renal White Paper: 59 statements, zero quality thresholds
  2. 2Odudu 2018The 139-427 cortical band and cortex-good/medulla-poor reliability split
  3. 3Harteveld 2020Ships the default outlier rule; proves labelling moves perfusion 1.8x
  4. 4Garcia-Ruiz 2025Only renal tSNR, PWS and cortex/medulla baselines at both field strengths
  5. 5Bones 2019Renal tSNR definition and values; tSNR below 1 is normal here
  6. 6Olsen 2025The 500 implausible-value ceiling, PET limits of agreement, unit trap
  7. 7Nery 2019Threshold-free weighted averaging as the alternative to hard rejection
  8. 8Wolf 2018Supplies the only hardcoded tissue constants: cortex/medulla T1

Read firstthe eight that decided the design

2 · Worked by hand

Worked by handevery formula, before any code

3 · Architecture & docs

Architecture & docsthe check map, then the document

4 · Live report

Live reportreal PASS output, not a mockup

Organ 3 of 3

Placenta 15 checks

placenta/PLACENTA_QC_DESIGN.md ↗

The thinnest evidence of the three, and the strictest rules because of it.

1 · Papers

  1. 1Taso 2023Proves no placental ASL consensus exists - the design's negative anchor
  2. 2Harteveld 2020The only reimplementable placental rules: outlier rejection, post-motion tSD, M0 scalar
  3. 3Zun 2017Healthy spatial CoV 0.58 kills the brain 0.67 cutoff; maternal position
  4. 4Abulnaga 2023Mask is graded, not trusted: human Dice 0.68, automated 82.8
  5. 5Uus 2020Only placenta-validated registration cut-offs: NCC 0.75 global, SSIM 0.6 local
  6. 6Dellschaft 2020Two-tier motion rejection; pump removes 40% volume, so nothing rigid-body
  7. 7Clement 2022ASL-BIDS is brain-only and excludes VSASL, so schema checks degrade
  8. 8Chernyavsky 2010Placentone radius ~2 cm: cotyledon heterogeneity is physiology, never a defect

Read firstthe eight that decided the design

2 · Worked by hand

Worked by handevery formula, before any code

3 · Architecture & docs

Architecture & docsthe check map, then the document

4 · Live report

Live reportreal PASS output, not a mockup

Still open

What happens next

  1. 1

    Still in review

    The checks are still going back and forth with mentors. As feedback arrives, more get added and existing ones corrected.

  2. 2

    More data

    Test against more datasets, especially real kidney and placenta data, which is the weakest part of the evidence today.

  3. 3

    Fix what breaks

    Bugs turn up when real people use it. I find them by running the tool, not by reading the diff, and fix them as they appear.

  4. 4

    Keep maintaining

    If more checks are needed I will keep adding them, and keep this repository maintained as the needs change.

Thank you

I did not come from a research background, or from ASL, or from biology. I came in knowing how to write software.

This summer taught me how to read a paper properly, how to tell a real threshold from a number somebody once picked, and how much of this field is still genuinely open.

That has turned into a real interest of its own, not just a way to finish the project.

Thank you for the trust and for the opportunity.

Agnik Misra · GSoC 2026 · OSIPI