Dr Amani Mwangi
102 m, the hospital

A second look at every slide.

I'm Dr Amani Mwangi, a pathologist and clinical AI lead in Nairobi. I build and validate AI that reads tissue slides, so district hospitals get a second look within days.

100 m, the lab

Five hospitals, two pathologists

In 2019 our unit at Tumaini Hospital read biopsies for five district hospitals with two pathologists. A result took three weeks, and in screening three weeks is a long time to wait.

10−2 m, the slide

Scanned where it is cut

Now each slide is scanned at the district hospital that prepared it. The image reaches Nairobi in minutes, where a courier used to take two days.

10−3 m, the tissue

Second Look reads the whole slide

The model my team built ranks every region of a slide by how much it needs a pathologist's eye, so the urgent cases are read first. It was trained on 38,000 slides from Kenyan hospitals, with consent and ethics approval.

10−5 m, the cells

Here, a person decides

Haematoxylin turns nuclei violet and eosin turns the rest pink. At this scale the decision belongs to a pathologist: the model flags, we read, and we sign. It has never signed a report.

The model's view

Where it looked, tested

An attention map shows where the model looked. It is not an explanation, so we test it: in our 2024 study, 97% of the slides it ranked urgent were confirmed urgent by two pathologists.

Scroll to zoom in

Research profile, not medical advice. The AI described supports pathologists and never replaces them.

Three studies, and what they changed

Every model we use is validated against pathologists before it touches a patient's result, and again at every new hospital.

2022
96.1%

Retrospective validation, 4,212 cervical biopsies

Sensitivity for high-grade lesions against the consensus of two pathologists, on slides from three hospitals the model had never seen.

2024
19 to 4days

Prospective study at five district hospitals

Median time from biopsy to signed report, after slides were scanned on site and ranked by Second Look before a pathologist read them.

2025
1,200slides

An open validation set for East Africa

Annotated slides released with consent so other teams can test their models on the stains and scanners used here.

Figures from the 2022 validation study.

A stained slide held in the clips of a microscope stage
On the stage, 10x
Amani and a lab technician reviewing a scanned slide at a district hospital
District hospital lab, 1x
Portrait of Dr Amani Mwangi in her lab
Dr Amani Mwangi, Nairobi

The record

  • 2019 to nowClinical AI lead, Tumaini Hospital NetworkFive district hospitals, one pathology unit in Nairobi
  • 2014 to 2019Consultant pathologist, Tumaini Hospital, Nairobi
  • 2021 to nowHonorary lecturerTeaches residents digital pathology at the multi-head microscope
  • 2018Fellowship in computational pathology, University of Leeds
  • 2008, 2014MBChB, then MMed in Pathology, University of Nairobi

Papers and talks

  • 2022A second look: validating slide triage in Kenyan hospitalsPaper
  • 2024Attention is not a diagnosisTalk to a pathology congress, Nairobi
  • 2025Scanning where the slide is cutPaper, the prospective study

Field 400 m, the hospital

Ask a question and the view moves to what the answer is about.

The view answers

Scripted answers from Amani's notes. Not Amani, and not medical advice. For anything else, propose a study and she replies herself.

Try asking

Fields of view

  1. Each answer you get is kept here. Select one to return to its view.

Propose a validation study

If your hospital, lab or ministry wants to test Second Look on its own slides, or to share data, tell me what you have in mind. I reply within five working days.

Kind of study
ReceivedSL-26-0000

Retrospective validation

Logged today. Amani replies within five working days.

Nothing is sent from this page; the proposal would reach her inbox on a live profile.