AI-assisted lesion risk stratification.
Upload a clinical or dermoscopic photograph. An EfficientNet B4·B5·B7 ensemble returns a probability distribution across nine dermatological conditions, with high-risk classes flagged for professional follow-up.
Submit a photograph for analysis.
Upload any skin lesion photograph. The ensemble returns a full probability distribution across all nine conditions — no account required.
Three-stage inference pipeline.
Image submission
Upload a clear, close-up photograph of the lesion. Higher resolution and controlled lighting improve classification accuracy.
Ensemble inference
EfficientNet B4, B5, and B7 each run an independent forward pass. Output probability vectors are averaged via soft voting.
Risk stratification
MEL, BCC, SCC, and AK predictions above threshold are flagged as high-risk and prompt a dermatologist referral recommendation.
Nine-class dermatological classification.
Training corpus: ISIC 2019, ISIC 2020, and HAM10000 — combined dermatoscopic and clinical image archives.
EfficientNet soft-vote ensemble.
Three EfficientNet variants are trained independently on the same corpus. At inference, per-class probability vectors are averaged element-wise — reducing single-model variance without retraining.
# soft-voting ensemble def predict_probs(self, img): per_model = [] for model in self.models: # B4, B5, B7 h, w = self.input_size(model) arr = _preprocess(img, (h, w)) per_model.append(model.predict(arr)[0]) return np.mean(per_model, axis=0) # soft vote # ImageNet normalization arr = (img / 255. - [.485,.456,.406]) / [.229,.224,.225]
Open to board-certified dermatologists.
We maintain a verified referral network for patients who receive high-risk screening results. If you're a practicing dermatologist, we'd like to list your practice.