AI Algorithms Catalogue
Legit.Health operates 59 distinct AI models in production, covering diagnosis support, severity quantification across 22+ clinical scales, wound assessment, surface segmentation, lesion detection, and phenotype classification. This catalogue summarises every model currently deployed in the platform. For clinical validation studies and peer-reviewed publications backing these models, see the Clinical Evidence page.
A few models serve more than one workflow (for example, Body Surface Segmentation supports both BSA scoring and wound assessment) and are counted once in the total of 59.
Several algorithms are new or newly validated since seed close (September 2024) and are marked with 🆕 across this catalogue: ALADIN (acne severity), AGPPGA (pustular psoriasis), APASI (psoriasis, automated PASI), Body Surface (surface quantification) and NMSC Malignancy detection (AUC 0.93). These represent the core of post-seed R&D investment.
1. Diagnosis Support
2. Severity Intensity Quantification (visual signs)
Ordinal classification on a 0-9 scale for 10 distinct clinical visual signs. These are the foundational building blocks of composite clinical scales (PASI, EASI, SCORAD, GPPGA, PPPASI, ODS).
| # | Visual sign | Category | Clinical scales supported |
|---|---|---|---|
| 1 | Erythema | Inflammatory | PASI · EASI · SCORAD · GPPGA · PPPASI |
| 2 | Desquamation | Morphological | PASI · GPPGA · PPPASI |
| 3 | Induration | Textural | PASI |
| 4 | Pustule | Lesion-type | PPPASI · GPPGA · Acne |
| 5 | Crusting | Morphological | EASI · SCORAD |
| 6 | Xerosis | Textural | EASI · SCORAD · ODS |
| 7 | Swelling | Inflammatory | EASI · SCORAD |
| 8 | Oozing | Inflammatory | EASI · SCORAD |
| 9 | Excoriation | Morphological | EASI · SCORAD |
| 10 | Lichenification | Textural | EASI · SCORAD |
Lead: Daniel Dagnino. Output format: ordinal 0-9 + continuous confidence score.
3. Wound Assessment (binary classification)
Wound care has been a major expansion area. 22 binary classifiers cover the AWOSI framework, NPUAP staging, TIME framework, and infection surveillance.
Wound edges (6 models)
| # | Model | Detects |
|---|---|---|
| 1 | Perilesional Erythema | Inflammation around wound |
| 2 | Damaged Edges | Compromised wound margins |
| 3 | Delimited Edges | Well-defined boundaries |
| 4 | Diffuse Edges | Poorly defined boundaries |
| 5 | Thickened Edges | Hyperkeratotic / rolled edges |
| 6 | Indistinguishable Edges | Severely compromised edges |
Exudate characterisation (5 models)
| # | Model | Detects |
|---|---|---|
| 7 | Perilesional Maceration | Moisture damage in periwound skin |
| 8 | Fibrinous Exudate | Normal healing response indicator |
| 9 | Purulent Exudate | Infection indicator |
| 10 | Bloody Exudate | Tissue fragility / hemorrhage |
| 11 | Serous Exudate | Clear/watery exudate |
Wound tissue assessment (11 models)
| # | Model | Detects |
|---|---|---|
| 12 | Biofilm-Compatible Tissue | Biofilm presence indicators |
| 13 | Affected Tissue: Bone | Bone involvement / osteomyelitis risk |
| 14 | Affected Tissue: Subcutaneous | Partial-thickness loss |
| 15 | Affected Tissue: Muscle | Muscle involvement |
| 16 | Affected Tissue: Intact Skin | Stage I pressure injury |
| 17 | Affected Tissue: Dermis-Epidermis | Partial-thickness skin loss |
| 18 | Wound Bed: Necrotic | Necrotic tissue presence |
| 19 | Wound Bed: Closed | Healed vs open |
| 20 | Wound Bed: Granulation | Healthy granulation |
| 21 | Wound Bed: Epithelial | Epithelialisation phase |
| 22 | Wound Bed: Slough | Devitalised tissue |
Lead: Daniel Dagnino. Use cases: AWOSI scoring, NPUAP staging, debridement planning, infection detection, healing prognosis.
4. Wound Staging & Composite Scores (2 models)
5. Surface Quantification (segmentation models, 13)
Pixel-level segmentation for surface area measurement. Outputs include percentage, absolute area (cm² when calibration available), and clinical alerts.
| # | Model | Application |
|---|---|---|
| 1 | 🆕 Body Surface Segmentation | 5-class (Head/Neck, Upper Extremities, Trunk, Lower Extremities) for BSA calculation. PASI/EASI/Burn assessment. |
| 2 | Erythema Surface | Wound erythema area + perilesional. Infection surveillance, AWOSI. |
| 3 | Wound Bed Surface | Total wound area + perimeter + max length/width. Healing rate. |
| 4 | Granulation Tissue Surface | Healing progression indicator. AWOSI, wound bed preparation. |
| 5 | Biofilm and Slough Surface | Debridement guidance. TIME framework. |
| 6 | Necrosis Surface | Urgent debridement indicator. Infection risk. |
| 7 | Maceration Surface | Moisture management. Dressing selection. |
| 8 | Orthopedic Material Surface | Exposed hardware detection. Surgical revision alerts. |
| 9 | Bone/Cartilage/Tendon Surface | Deep structure exposure. Osteomyelitis risk. Amputation risk. |
| 10 | Hair Loss Surface | 3-class (Hair, No Hair, Non-Scalp). SALT, APULSI scoring. |
| 11 | Nail Lesion Surface | NAPSI, OSI scoring. |
| 12 | Hypopigmentation / Depigmentation Surface | VASI, VETF. Vitiligo assessment. |
| 13 | Hyperpigmentation Surface | MASI, mMASI. Melasma + PIH assessment. |
Lead: Daniel Dagnino (most) + Ignacio Hernández (nail).
6. Lesion Detection (object detection, 5 models)
Object detection (YOLO architecture) for individual lesion counting and localisation. Bounding boxes + confidence scores.
7. Phenotype & Pattern Classification (2 models)
8. Operational / Non-clinical Models (5)
Pipeline and quality assurance models that enable the rest of the platform to operate at clinical-grade reliability.
| # | Model | Function |
|---|---|---|
| 1 | DIQA (Dermatology Image Quality Assessment) | 0-10 quality score with dimension subscores. Telemedicine + quality control. Published JAAD 2023. |
| 2 | Domain Validation | Multi-class: non-skin / skin clinical / skin dermoscopic. Image routing. |
| 3 | Skin Surface Segmentation | Binary skin region detection. Preprocessing, ROI extraction. |
| 4 | Body Surface Segmentation | 5-class anatomical regions. BSA calculation (clinical use, listed in §5 above). |
| 5 | Head Detection | Privacy protection + patient counting + multi-patient flag. |
Leads: Ignacio Hernández (DIQA, Domain Validation), Alberto Sabater (Skin Surface, Head Detection).
Strategic implications for investors
| Competitor | Approximate model count | Coverage |
|---|---|---|
| Legit.Health | 59 | Multi-pathology platform · 22+ clinical scales |
| Skin Analytics | ~1 | Melanoma detection only |
| DermaSensor (FDA cleared) | ~1 | Skin cancer detection only |
| SkinVision | 1-2 | Skin cancer self-screening |
| MetaOptima (DermEngine) | 3-5 | Diagnosis support + analysis |
Building a multi-pathology platform with 59 distinct models requires 3-5 years of catch-up investment in data, annotation, and clinical validation. This is the durable technical moat that justifies higher valuation multiples vs. single-purpose competitors.
Pipeline coverage map by clinical area
| Therapeutic area | Models supporting this area | Clinical scales |
|---|---|---|
| Psoriasis | 4 (intensity) + 1 segmentation | PASI · PPPASI · GPPGA |
| Atopic dermatitis | 7 (intensity) + 1 segmentation | EASI · SCORAD · ODS |
| Hidradenitis suppurativa | 1 detection (4 classes) + 2 phenotype | IHS4 · Martorell · Hurley · HS-PGA |
| Acne | 2 detection + 1 intensity | GAGS · IGA · ALADIN |
| Urticaria | 1 detection | UAS7 · UCT · AUAS |
| Wound care | 22 binary + 1 stage + 1 composite + 7 segmentation | AWOSI · NPUAP · TIME framework |
| Alopecia | 1 detection + 1 segmentation | SALT · APULSI |
| Nail diseases | 1 segmentation | NAPSI · OSI |
| Vitiligo | 1 segmentation | VASI · VETF |
| Melasma / PIH | 1 segmentation | MASI · mMASI |
| Skin cancer | ICD diagnosis + binary indicators | Triage urgency |
| Quality control | DIQA · Domain · Skin Surface · Head Detection | - |
See also
- Clinical Evidence: 9 peer-reviewed papers + 7-paper pipeline validating these algorithms
- Medical Device Certification: regulatory framework for these models
- Commercial Metrics: pharma + clinical use cases enabled by each algorithm