Cancer3.AI › AI in Oncology › AI Models › Mirai
Task-specific model Open weights Research use only radiology

Mirai

Mammography-based model predicting five-year breast cancer risk from the four standard screening views, designed to be robust across hospitals and to work with or without clinical risk factors.

GitHub Paper DOI { } JSON

At a glance

DeveloperMIT Jameel Clinic / Massachusetts General Hospital
Released2021-01-27
LicenceMIT
AvailabilityOpen weights
KindTask-specific model
Regulatory statusResearch use only

What it does

Outputs risk at years 1–5 from a screening mammogram; the paper compares it with the Tyrer-Cuzick model and reports higher discrimination on internal and external cohorts.

Tasks, data types and cancers

TaskRisk predictionScreening
Data typeMammography
Clinical settingScreening
CancerInvasive Breast Carcinoma
Inputfour-view screening mammogram (+ optional risk factors)
Outputrisk at years 1–5

Architecture

Familyconvolutional image encoder + transformer over views
Pre-trainingsupervised on MGH screening mammograms with follow-up

Training data

Screening mammograms from Massachusetts General Hospital with cancer follow-up; external validation at Karolinska (Sweden) and Chang Gung Memorial Hospital (Taiwan).

InstitutionsMGH (USA); Karolinska (SE); CGMH (TW)

Evaluation

Benchmark / datasetMetricValueExternal validationSource
MGH test / Karolinska / CGMH C-index (5-year risk) higher than Tyrer-Cuzick v8 in all three (paper) yes Source

How to run

LibraryPyTorch (repo)

Weights and inference code on GitHub; requires DICOM preprocessing described in the repo.

Regulatory status and intended use

Regulatory statusResearch use only
Intended useResearch.

Source →

Regulatory status is quoted from the source linked above and can change. Research-use-only models must not be used for clinical decisions.

Limitations and bias

  • Vendor-specific mammography systems in training; calibration must be checked per site.
  • Risk model, not a lesion detector.

Sources

  1. Yala A et al. Toward robust mammography-based models for breast cancer risk. Sci Transl Med 2021

This page is educational — it is not medical advice and does not replace consultation with an oncologist. Diagnostic and treatment decisions are made solely by specialist physicians.