SecondLook
Live, open-source demo

The detector spots basal cell carcinoma.
The trust layer makes the call worth trusting.

SecondLook reads skin-pathology slides for basal cell carcinoma with a detector built on 129,364 labeled tiles. On every confident call it runs a spellcheck - checking where the model looked - so you trust the reads that pass and take a second look at the ones that don't.

Research demo not a medical device not for diagnosis
After correction: Grad-CAM heat sits on tissue across the field Before correction: Grad-CAM heat pooled in the top-right padding corner before · corner heat after · on tissue
FLAGGED Claude
Attention check
Score is maxed at 1.00, but the heat sits on a padding corner, not tissue.
Corner
Heat
1.00
Score

Same tile. Same 1.00 confidence. Drag to compare where each model looked.

VERIFIED

Attention sits on tissue. The check passed and the finding stands.

FLAGGED

High score, but the heat is on a corner or edge. Routed for human review.

DEFER

Evidence is unclear, or the check fails closed. No call is made.

01 The Problem

A model can be confidently wrong.

Two detectors can post the same 1.00 score on the same tile. One read the lesion. The other locked onto a padding corner and never saw it.

The metrics can't tell them apart. The attention map can.

02 How It Works

A spellcheck for the model's work, never the slide.

On a confident positive, the trust layer reads the score, the attention, and the heatmap - then returns a verdict in plain language. The detector's call never changes; the doctor keeps the final say.

03 The Evidence, in One Tile

Same tile. Same confidence. Opposite trust.

Two EfficientNet-B4 models, identical but for one training change. Before learned a shortcut and stared at a corner. After reads tissue. Aggregate metrics stay high on both - only the attention map separates them.

Original H&E tile
Before corner heat → FLAGGED
After tissue heat → VERIFIED
Three panels: original H&E tile, before model with heat pooled in a corner, after model with heat spread across tissue
Same detector class, same 1.00 scoreOnly the attention map tells them apart
04 Evidence & Standing

Built on a real dataset, shown in the open.

0.000
Test AUC
0
Held-out test tiles, patient-disjoint split
3-tier
Positive, uncertain, or negative at 0.7 / 0.3
0
Labeled tiles, Heidelberg (CC BY)
Associated Research
AAD Innovation Academy 2026 · Poster #85613
Beyond Binary Classification - a three-tier BCC framework built on the same detector.
Anthropic
Built with Claude · Life Sciences
Selected among 500 of 6,000+ builders to compete in the hackathon.
Clinical Advisor
Henry Lim, D.O.
Chief Dermatology Resident Physician.

Metrics are tile-level, on an internal held-out split of one dataset (Heidelberg), patient-disjoint.

05 Run It Yourself

The demo is hosted. The whole thing is open.

Both checkpoints ship in the repo, so you can reproduce the before / after attention without retraining. Claude's check calls the Anthropic API with your own key, kept server-side.

Apache-2.0 Weights via Git LFS Runs on CPU or Apple Silicon

See where it looked.

Open Compare on a positive tile and watch corner heat become FLAGGED and tissue heat become VERIFIED - same detector, different trust.