Case identity and source
- Slide
- De-identified WSI identifier
- Source
- Originating lab or repository
- Sample
- Specimen and block relationship
The TrueDx evidence-rich report
A pathology opinion you can see—not only read.
Representative slide regions, annotations, maps, and measurements connect the written interpretation to visual evidence patients can understand and oncologists can inspect. The reviewing pathologist verifies the relevant findings and remains the diagnostic authority who signs the final report.
Image fitness and scale
- Dimensions
- 21,911 × 21,753 px
- Resolution
- 0.2527 µm/px · 40×
- Physical extent
- 5.54 × 5.50 mm
Exactly what was analyzed
- Tissue map
- 43 cores · 4.229 mm²
- Nuclei map
- 62 cores · 16.61 mm²
- MIL subtype
- 325 patches · 20×
What “evidence-rich” means
More than a narrative report.
A conventional report records the pathologist's conclusion. An evidence-rich report also presents the computational and visual observations behind that conclusion in a form the care team can inspect.
1Model-derived evidencePredictions, maps, counts, measurements, and representative fields.
2Pathologist interpretationMicroscopic findings, diagnostic reasoning, limitations, and final impression.
3Traceable provenanceModel version, parameters, tissue coverage, artifact lineage, and reviewed regions.
Six evidence layers
One slide, examined from complementary angles.
Each layer answers a different question. None should be treated as a stand-alone diagnosis.
Prediction heatmap
Shows where a classification model found evidence supporting—or arguing against—a predicted category across slide patches.
- Predicted class and calibrated score
- Spatial attention or evidence map
- Highest-support and strongest-counterexample fields
Nuclei map
Segments individual nuclei and assigns model-derived morphological classes across evaluated tissue regions.
- Total nuclei and analyzed area
- Neoplastic, stromal, inflammatory and other classes
- Counts, proportions, density and heterogeneity
Tissue map
Separates tissue architecture into computational regions such as tumor, stroma, inflammation, necrosis, and other tissue.
- Area fractions and coverage
- Tumor–stroma ratio
- Regional distribution and representative fields
Biomarker map
Displays predicted marker intensities or phenotypes from a validated in-silico model when that analysis is available.
- Marker or phenotype distribution
- Cell counts and regional co-localization
- Agreement and divergence from morphology
Microscopic findings
Converts relevant visual and quantitative observations into reviewable pathology findings.
- Cellular composition
- Tumor-infiltrating lymphocytes
- Necrosis and tissue architecture
Artifact provenance
Records how every model-derived number and image was produced so it can be traced and audited.
- Model name, version and parameters
- Analyzed area, tiles, magnification and resolution
- Artifact dependencies and source regions
Interactive evidence comparison
Move the slider from H&E to model overlay.
Each comparison keeps the original field and generated evidence in the same coordinate system. Drag horizontally to inspect exactly what the model added.
Representative field supporting a stroma-predominant pattern.
Field selected for the densest lymphocyte and plasma-cell evidence.
Best-supported necrosis field among adequately filled analyzed tiles.
Patch carrying strong model evidence for the candidate subtype.
Region where morphology and predicted phenotype show close agreement.
Images are extracted from the supplied de-identified demonstration report. Overlays are model-derived artifacts for pathologist review and are not independent diagnoses.
Ancillary studies
Model-derived studies that extend the microscopic review.
These analyses add structured computational evidence to the case. They are reported separately from the pathologist's final diagnosis and include their own methods, coverage, uncertainty, and limitations.
Study 01Model-derived
Histological subtype — attention-based MIL
An attention-based multiple-instance model over CONCH v1 patch features evaluates the distribution of evidence for candidate histological subtypes.
325 patches · 512 px · level 0 · 20× sampling
p(IDC)0.8216
p(ILC)0.1784
Top-10 attention mass24.2%
- Shows how evidence is distributed across slide patches
- Identifies the strongest supporting patch
- Surfaces the strongest field arguing against the prediction
- Requires confirmation by direct microscopic review
The probability is a model output, not a stand-alone subtype diagnosis.
Study 02Virtual phenotype
In-silico immunophenotype
A virtual multiplex-immunofluorescence model predicts a 20-marker panel from H&E and maps predicted phenotypes back to previously segmented nuclei.
GigaTIME-Flash · 43 cores · 11.517 mm² · 16,084 cells
CKCD3CD138CD68CD34
- Reports predicted phenotype counts and proportions
- Compares virtual phenotype with morphological classes
- Highlights regions of agreement and divergence
- Preserves model version, analyzed area, and coverage
No antibody was applied to the analyzed section. These are predicted marker intensities—not performed stains and not a replacement for validated IHC.
Illustrative report anatomy
From computational evidence to signed opinion.
The values below are a fictional presentation example—not a patient result or a claim of model performance.
TrueDx second-opinion consultationIllustrative breast core biopsy
DEMOPathologist impression
Invasive carcinoma is identified in the reviewed material.
Final subtype, grade, and biomarker interpretation require direct pathologist review and correlation with validated studies.
Evidence package7traceable artifacts
Tissue architectureComputed + reviewed
Stroma65.7%
Tumor15.9%
Necrosis2.9%
Tumor–stroma ratio0.80
Stroma-predominant field with focal tumor regions and patchy necrosis.
Cellular compositionComputed + reviewed
Total nuclei evaluated16,093
Tumor nuclei23.4%
Stromal nuclei74.7%
Analyzed area16.61 mm²
Marked regional heterogeneity is present; field-level values accompany the slide map.
Tumor-infiltrating lymphocytesComputed + reviewed
Lymphocytes + plasma cells1.9%
Peak regional fraction4.8%
PatternStromal
The illustrative field is lymphocyte-poor; representative support and counterexample regions are retained.
NecrosisComputed + reviewed
Classified tissue area2.9%
DistributionFocal
PatternPatchy
The most supportive region and a low-confidence counterexample are shown for pathologist verification.
SupportsRepresentative positive field
The image region that most strongly supports the stated finding.
ChallengesCounterexample field
The region that least supports or most strongly challenges the stated finding.
ResolvesPathologist judgment
The reviewer determines what is diagnostically valid and what must be excluded.
Homegrown models + Agentic Force
Generated, extracted, organized—then clinically reviewed.
Path23's homegrown AI models generate spatial predictions and extract tissue and cellular measurements. Agentic Force coordinates the artifacts, selects traceable supporting and opposing fields, drafts structured findings, and assembles them for review.
The pathologist remains the diagnostic authority.
- AI outputs are clearly labeled as model-derived
- Predicted biomarkers are not represented as performed stains
- Uncertainty, coverage, and limitations remain visible
- The pathologist may accept, modify, or reject every generated statement
- Only the reviewed and signed report becomes the clinical opinion
Commercial demonstrations
See an evidence-rich report on an appropriate demonstration case.
Discuss laboratory onboarding, supported analyses, validation status, and report configuration with TrueDx.
Request a Path23 demonstration
Compose a demonstration enquiryDo not email patient information or slide files.