Classification overview
The model outputs a probability distribution across seven classes, each representing a specific type of skin lesion:Malignant lesions
Cancerous or pre-cancerous conditions requiring medical attention:
- Actinic Keratoses
- Basal Cell Carcinoma
- Melanoma
Benign lesions
Non-cancerous conditions with lower clinical urgency:
- Benign Keratoses
- Dermatofibroma
- Melanocytic Nevus
- Vascular Lesion
While the model categorizes lesions as malignant or benign, all predictions should be verified by a qualified dermatologist. This tool is designed to assist screening, not replace professional medical diagnosis.
The seven classification categories
1. Actinic Keratoses (akiec)
Class index: 0 Medical classification: Pre-cancerous lesion Description: Actinic keratoses are rough, scaly patches on the skin caused by years of sun exposure. They are considered pre-cancerous because a small percentage can progress to squamous cell carcinoma if left untreated.Clinical characteristics
Clinical characteristics
Appearance:
- Rough, dry, or scaly patches
- Flat to slightly raised
- Color ranges from pink to red to brown
- Typically 1-3 cm in diameter
- Texture often described as feeling like sandpaper
- Face and ears
- Scalp (especially in balding areas)
- Backs of hands
- Forearms and shoulders
- Any sun-exposed area
Risk factors and prevalence
Risk factors and prevalence
High-risk populations:
- Fair-skinned individuals
- People over 40 years old
- Those with significant sun exposure history
- Immunosuppressed patients
Treatment and prognosis
Treatment and prognosis
Treatment options:
- Cryotherapy (freezing)
- Topical medications (5-fluorouracil, imiquimod)
- Photodynamic therapy
- Curettage or laser therapy
2. Basal Cell Carcinoma (bcc)
Class index: 1 Medical classification: Malignant (cancerous) but rarely metastatic Description: Basal cell carcinoma is the most common form of skin cancer. It develops in the basal cells of the skin’s epidermis and rarely spreads to other parts of the body, but can cause significant local damage if not treated.Clinical characteristics
Clinical characteristics
Appearance:
- Pearly or waxy bump
- Flat, flesh-colored or brown scar-like lesion
- Bleeding or oozing sore that heals and returns
- Pink growth with raised, rolled border and central depression
- Nodular (most common): Pearly, dome-shaped bump
- Superficial: Red, scaly patch
- Morpheaform: Scar-like, firm lesion
- Pigmented: Brown, blue, or black appearance
Risk factors and prevalence
Risk factors and prevalence
High-risk factors:
- Chronic sun exposure
- Fair skin, blonde or red hair, light-colored eyes
- History of sunburns, especially in childhood
- Previous skin cancer
- Radiation therapy
- Weakened immune system
Treatment and prognosis
Treatment and prognosis
Treatment options:
- Surgical excision
- Mohs micrographic surgery (for high-risk areas)
- Curettage and electrodesiccation
- Radiation therapy
- Topical treatments for superficial types
3. Benign Keratoses (bkl)
Class index: 2 Medical classification: Benign (non-cancerous) Description: Benign keratoses include seborrheic keratoses and other benign skin growths. These are harmless, non-cancerous growths that appear with aging and do not require treatment unless they cause discomfort or cosmetic concerns.Clinical characteristics
Clinical characteristics
Appearance:
- Brown, black, or tan growths
- Waxy, slightly raised appearance
- “Stuck-on” look, as if painted onto the skin
- Round or oval shape
- Texture can be smooth or rough
- Size varies from very small to over 1 inch
- Seborrheic keratoses: Most common type
- Solar lentigo: Flat, brown spots from sun exposure
- Lichenoid keratosis: Inflamed variant, may be red or purple
Risk factors and prevalence
Risk factors and prevalence
Occurrence patterns:
- Increases with age (most common after 50)
- Genetic predisposition
- No direct relationship with sun exposure
- Can appear anywhere on the body except palms and soles
Treatment and considerations
Treatment and considerations
When to treat:
- Cosmetic concerns
- Irritation from clothing or jewelry
- Itching or bleeding
- Diagnostic uncertainty (to rule out melanoma)
- Cryotherapy
- Curettage
- Electrocautery
- Laser therapy
4. Dermatofibroma (df)
Class index: 3 Medical classification: Benign (non-cancerous) Description: Dermatofibromas are common, benign skin nodules composed of fibrous tissue. They are typically firm to the touch and often develop after minor skin trauma, such as insect bites or small injuries.Clinical characteristics
Clinical characteristics
Appearance:
- Firm, raised bump
- Color ranges from pink to red-brown to dark brown
- Usually 0.5-1 cm in diameter
- May feel like a hard marble under the skin
- “Dimple sign”: Pinching causes dimpling of the center
- Hard or rubbery consistency
- May be slightly tender when pressed
- Generally asymptomatic
Development and prevalence
Development and prevalence
Common triggers:
- Insect bites
- Minor skin trauma
- Splinters or thorns
- Folliculitis
- More common in women
- Typically appears in young to middle-aged adults
- Usually found on legs, but can occur anywhere
Management
Management
Observation:
- No treatment necessary for typical dermatofibromas
- Benign with no cancer risk
- Does not require biopsy if diagnosis is confident
- Requested for cosmetic reasons
- Causes discomfort or repeated irritation
- Diagnosis is uncertain
- Surgical excision (may leave scar)
- Cryotherapy (partial removal)
- Laser therapy
5. Melanoma (mel)
Class index: 4 Medical classification: Malignant (cancerous) with metastatic potential Description: Melanoma is the most dangerous form of skin cancer. It develops in melanocytes (pigment-producing cells) and has a high potential to spread to other organs if not detected and treated early.Clinical characteristics (ABCDE rule)
Clinical characteristics (ABCDE rule)
A - Asymmetry: One half doesn’t match the other halfB - Border irregularity: Edges are ragged, notched, or blurredC - Color variation: Multiple colors (brown, black, tan, red, white, blue)D - Diameter: Larger than 6mm (pencil eraser size), though can be smallerE - Evolving: Changing in size, shape, color, or elevation; new symptoms like bleeding or itchingAdditional warning signs:
- Sore that doesn’t heal
- Spread of pigment beyond border
- Redness or swelling
- Itching, tenderness, or pain
Types of melanoma
Types of melanoma
Superficial spreading melanoma (70%):
- Most common type
- Grows horizontally before invading deeper
- Can arise in existing mole
- Aggressive form
- Grows vertically into skin
- Often appears as dark bump
- Develops in sun-damaged skin
- Common in elderly on face and arms
- Occurs on palms, soles, under nails
- More common in darker skin tones
Risk factors and statistics
Risk factors and statistics
Major risk factors:
- Personal or family history of melanoma
- Many moles (over 50) or atypical moles
- Fair skin, light hair, light eyes
- Severe sunburns, especially in childhood
- Tanning bed use
- Immunosuppression
- Older age (risk increases with age)
- Over 100,000 new cases annually in the United States
- Accounts for 1% of skin cancers but majority of skin cancer deaths
- 5-year survival rate: 99% if detected early (localized), 27% if spread to distant organs
Treatment and prognosis
Treatment and prognosis
Staging determines treatment:Stage 0-II (localized):
- Surgical excision with margins
- Sentinel lymph node biopsy for thicker lesions
- Wide excision
- Lymph node dissection
- Immunotherapy or targeted therapy
- Systemic therapy (immunotherapy, targeted therapy)
- Radiation for symptom control
- Clinical trials
- Breslow thickness (depth of invasion)
- Ulceration
- Mitotic rate
- Lymph node involvement
- Distant metastasis
Melanoma is the most critical classification for this model to identify accurately. Any suspicious lesion flagged as possible melanoma should receive immediate dermatological evaluation.
6. Melanocytic Nevus (nv)
Class index: 5 Medical classification: Benign (non-cancerous) Description: Melanocytic nevi, commonly called moles, are benign proliferations of melanocytes (pigment cells). While benign, monitoring is important as melanoma can occasionally arise from pre-existing moles.Clinical characteristics
Clinical characteristics
Typical appearance:
- Round or oval
- Uniform brown, tan, or pink color
- Clearly defined borders
- Flat or slightly raised
- Diameter typically less than 6mm
- Symmetrical appearance
- Size varies from small to giant
- Slightly higher melanoma risk for large variants
- Most common type
- Peak numbers in 20s-30s, then gradually fade
- Irregular borders
- Variable coloring
- Larger than typical moles
- Higher melanoma risk marker
Development and lifecycle
Development and lifecycle
Growth pattern:
- Appear during childhood and adolescence
- Peak mole count in 20s-40s
- Gradually fade or disappear in older age
- Most adults have 10-40 moles
- Junctional nevus (flat, at dermal-epidermal junction)
- Compound nevus (slightly raised)
- Intradermal nevus (dome-shaped, flesh-colored)
When to be concerned
When to be concerned
Normal mole characteristics:
- Appears before age 30
- Symmetrical
- Uniform color
- Stable over time
- Similar to other moles on body
- New mole after age 30
- Changes in size, shape, or color
- ABCDE criteria present
- Itching, bleeding, or crusting
- Looks different from other moles (“ugly duckling sign”)
Management and treatment
Management and treatment
Observation: Most moles require no treatmentRemoval indications:
- Suspicious features warranting biopsy
- Cosmetic concerns
- Repeated irritation from clothing or jewelry
- Patient anxiety
- Excisional biopsy (complete removal with margin)
- Shave excision (for clearly benign, raised moles)
- Pathological examination to confirm benign nature
Distinguishing between benign melanocytic nevi and melanoma is one of the most challenging aspects of dermatological AI. The model may occasionally confuse these categories due to their visual similarity.
7. Vascular Lesion (vasc)
Class index: 6 Medical classification: Benign (non-cancerous) Description: Vascular lesions are benign skin conditions caused by abnormal blood vessels. This category includes various types such as cherry angiomas, spider angiomas, and other vascular birthmarks.Clinical characteristics
Clinical characteristics
General appearance:
- Red, pink, or purple discoloration
- May be flat or raised
- Blanch (turn white) when pressed
- Size varies widely
- Bright red, dome-shaped bumps
- 1-5mm in diameter
- Increase with age
- Commonly on trunk
- Central red spot with radiating vessels
- Blanches from center outward
- Common on face, neck, upper trunk
- Flat, pink to red birthmarks
- Present from birth
- Gradually darken and thicken with age
Causes and development
Causes and development
Cherry angiomas:
- Benign proliferation of capillaries
- Increase in number with age
- Genetics play a role
- Nearly universal in people over 70
- Associated with pregnancy, liver disease, or normal occurrence
- May resolve spontaneously
- Increased estrogen levels contribute
- Congenital capillary malformation
- Present at birth
- Do not resolve spontaneously
- May be associated with syndromes (e.g., Sturge-Weber)
Medical significance
Medical significance
Generally benign: No cancer riskWhen to investigate further:
- Sudden appearance of multiple spider angiomas (may indicate liver disease)
- Bleeding or painful vascular lesions
- Rapid growth or change
- Located in cosmetically sensitive area
- Spider angiomas: Liver cirrhosis, pregnancy, hyperthyroidism
- Multiple cherry angiomas: Normal aging; very rarely associated with systemic disease
Treatment options
Treatment options
Indications for treatment:
- Cosmetic concerns
- Bleeding (especially for traumatized lesions)
- Patient preference
- Laser therapy (pulsed dye laser most effective)
- Electrocautery
- Cryotherapy
- Surgical excision (rarely needed)
Model output interpretation
The model outputs seven probability scores (one per class) that sum to 1.0:In this example, the model predicts Melanoma with 62% confidence, followed by Basal Cell Carcinoma at 12%. The user should be advised to consult a dermatologist immediately given the high melanoma probability.
Classification confidence thresholds
When interpreting results, consider implementing confidence thresholds:| Confidence Level | Recommendation |
|---|---|
| Above 80% | High confidence prediction; still recommend professional verification |
| 50-80% | Moderate confidence; definitely consult dermatologist |
| 30-50% | Low confidence; multiple possibilities, professional evaluation essential |
| Below 30% | Very uncertain; model cannot reliably classify, seek expert opinion |
Regardless of confidence level, this model should never be used as a replacement for professional medical diagnosis. All concerning lesions should be evaluated by a qualified dermatologist.
Common classification challenges
Melanoma vs. melanocytic nevus
The most critical challenge. Both are pigmented lesions arising from melanocytes. ABCDE criteria help distinguish, but subtle cases require expert evaluation.
Actinic keratoses vs. benign keratoses
Can appear similar, but actinic keratoses are rougher, occur in sun-exposed areas, and have malignant potential.
Basal cell carcinoma variants
Multiple subtypes (nodular, superficial, pigmented) can mimic other lesions like melanoma or benign growths.
Image quality impact
Lighting, focus, and image angle significantly affect classification accuracy. Clear, well-lit images yield best results.
Clinical decision support
This classification system is designed as a screening tool, not a diagnostic device:Appropriate uses
- Initial triage of concerning lesions
- Educational tool for understanding skin cancer types
- Encouraging individuals to seek professional evaluation
- Monitoring lesion changes over time (with photos)
Inappropriate uses
- Definitive medical diagnosis
- Treatment planning without biopsy confirmation
- Avoiding dermatologist visits for suspicious lesions
- Self-diagnosis of skin cancer
Regulatory consideration: This model is not FDA-approved as a medical device and should not be marketed or used as a substitute for professional dermatological evaluation.
Next steps
Model architecture
Understand the CNN architecture that performs classification
Training process
Learn how the model was trained on dermatological images