Research & Validation
We publish our methods, share our datasets and submit every model to rigorous external validation before it reaches a patient.
95%
Classification accuracy · ISIC held-out validation
±0.18 mm
Depth MAE · Breslow regression
140+
Skin conditions recognised from phone photo
12
Peer-reviewed publications in submission
Validation pipeline
Four phases from lab bench to bedside. No shortcuts.
Internal validation
Stratified k-fold cross-validation on ISIC 2016–2024, HAM10000 and our proprietary DermaPrime Atlas. Class-balanced sampling and demographic parity checks.
External validation
Independent test sets from Cracow Medical Hospitals and Stanford Dermatology. Images captured on different devices, by different operators, in different lighting.
Prospective clinical study
A multi-centre, prospective reader study comparing DermaPrime against board-certified dermatologists on consecutive real-world patient images. Primary endpoint: sensitivity & specificity.
Regulatory submission
CE-IVDR technical documentation and FDA 510(k) pre-submission meetings. Expected regulatory clearance timeline: Q4 2026.
Peer-reviewed research
Selected melanoma & dermoscopy publications by our CEO and scientific lead, Prof. Joanna Jaworek-Korjakowska, and collaborators. Full bibliography on Google Scholar.
Melanoma Thickness Prediction Based on Convolutional Neural Network with VGG-19 Model Transfer Learning
Jaworek-Korjakowska J., Kleczek P., Gorgon M. · IEEE/CVF CVPR Workshops (ISIC Skin Image Analysis)
Analysis of the ISIC image datasets: Usage, benchmarks and recommendations
Cassidy B., Kendrick C., Brodzicki A., Jaworek-Korjakowska J., Yap M. H. · Medical Image Analysis, vol. 75
Interpretability of a deep learning based approach for the segmentation of skin lesions in dermoscopic images
Jaworek-Korjakowska J., Brodzicki A., Cassidy B., Kendrick C., Yap M. H. · Sensors / IEEE Access
Skin_Hair Dataset: Setting the Benchmark for Effective Hair Inpainting Methods for Improving the Image Quality of Dermoscopic Images
Jaworek-Korjakowska J., Wójcicka A., Kucharski D., Brodzicki A., Kendrick C., Cassidy B., Yap M. H. · ECCV Workshops
Automatic Classification of Specific Melanocytic Lesions Using Artificial Intelligence
Jaworek-Korjakowska J., Kłeczek P. · BioMed Research International
A Deep Learning Approach to Vascular Structure Segmentation in Dermoscopy Colour Images
Jaworek-Korjakowska J. · BioMed Research International
Computer-Aided Diagnosis of Micro-Malignant Melanoma Lesions Applying Support Vector Machines
Jaworek-Korjakowska J. · BioMed Research International
eRDA: Efficient Reasoning for Melanoma Detection with Augmented Decision Aids
Brodzicki A., Jaworek-Korjakowska J., et al. · Pattern Recognition Letters
Data transparency
We test on public datasets, but we train on our own datasets. Our models are trained on a carefully curated mix of public dermatology archives and proprietary clinical datasets. We disclose provenance, bias audits and exclusion criteria.
- ISIC Archive (2016–2024) — 400,000+ dermoscopic images
- HAM10000 — 10,015 dermatoscopic images of common pigmented lesions
- DermaPrime Atlas — 50,000+ smartphone & dermoscope images with histopathology correlation
- Annual bias audits by independent epidemiologists
- Demographic parity checks across Fitzpatrick skin types I–VI
Ethical data pledge
We never purchase scraped or unconsented medical images. Every image in the DermaPrime Atlas was captured with informed consent, anonymised at source, and is governed by a Data Processing Agreement aligned with GDPR and HIPAA.
We publish annual transparency reports on model performance across skin-tone groups, age bands and anatomic sites.
Interested in collaborating?
We are open to academic partnerships, multi-centre studies and independent model audits.