Nemo · IHCBuilt on Nemo

Nemo PD-L1

PD-L1 scoring on IHC

PD-L1 scoring across assays: tumour proportion score and combined positive score from per-cell classification.

Nemo PD-L1 output — PD-L1 IHC at 40×: positive (green) and negative (red) cells · drag to compare

Nemo PD-L1: raw slide — PD-L1 IHC at 40×: positive (green) and negative (red) cells · drag to compare

The problem

PD-L1 scoring varies between readers

TPS and CPS require counting positive tumour cells and positive immune cells against all viable tumour cells — by eye, across a whole tumour, to a threshold of 1 % or 50 %.

Why it matters

Treatment eligibility depends on the threshold

Immunotherapy eligibility turns on TPS ≥ 1 % or ≥ 50 %, or CPS ≥ 1, 10 or 20. A count places the case on the right side of the line for a reason.

Two scores, one slide

Which cells count, and when

TPS counts PD-L1-positive tumour cells only. CPS adds the positive immune cells. Switch between them to see which cells go into each score.

PD-L1+ tumour cells ÷ viable tumour cells × 100 %

PD-L1 IHC with every cell called positive or negative by Nemo

TPS · tumour cells onlyCPS · tumour + immune cells

PD-L1 positivePD-L1 negative

What changes

Three things you stop doing by hand

01
Tumour and immune, separated
Tumour and immune cells classified and scored for PD-L1 positivity individually.
02
Same score at every site
Same TPS and CPS for the same slide, every read.
03
Both scores, one slide
Whole-tumour scoring rather than an estimate over selected fields.

Outputs

What you get back

TPSpositive tumour cells / viable tumour cells, with the threshold applieddata

CPS(positive tumour + immune cells) / viable tumour cells, capped at 100data

Cell maptumour and immune cells, positive and negativeoverlay

Per-cell exportclass and positivityCSV / JSON

Reported to standard

TPS and CPS calculated per the assay interpretation manual (e.g. PD-L1 IHC 22C3 pharmDx) · Reference →

Evidence

Validation

Cell detection, per-cell status and TPS on held-out regions.

Cell detection · 32 held-out regions

F1 0.82, just below human agreement

Nemo 0.82Pathologists 0.87

0.40.50.60.70.80.91.0

Nemo vs each of 4 pathologists, 0.80–0.83Inter-pathologist agreement

Precision0.85

Recall0.78

F10.82

Four pathologists annotated the same regions independently. Nemo agrees with each of them at 0.80–0.83, so it is not fitted to one reader’s habits.

PD-L1 status, per cell

89% agreement with expert calls

89%agreement

0.96AUC

0.86sensitivity

0.92specificity

5,552 tumour cells across 98 regions.

TPS, per region · 23 regions with ≥100 tumour cells

r = 0.93 with reference TPS, no systematic bias

78%

TPS ≥ 1%✓ none missed

96%

TPS ≥ 5%✓ none missed

87%

TPS ≥ 10%✓ none missed

91%

TPS ≥ 50%

Agreement on which side of each cut-off a region falls. At 1%, 5% and 10% no region above the threshold was missed (sensitivity 100%); at 1%, specificity is 0.375, so some negative regions are over-called.

0bias +1.1+34.1−31.9

Bland–Altman summary: Nemo − reference TPS, in points. Shaded: 95% limits of agreement.

0.93Pearson r

+1.1bias, points

9.8MAE, points

±3395% limits

Validation scope

2 assays: PD-L1 E1L3N and SP263
2 centres, 2 scanner vendors: 3DHISTECH Pannoramic, Roche Ventana iScan
Public dataset, held-out test split, 4 independent annotators
Region-level (~1.5 mm), not whole-slide

All Nemo validation results →

Also in the Nemo family

Nemo

Try Nemo PD-L1

Run it on a slide in the platform, or ask for the product sheet with the full output specification.

See it in your slide Request a demo