Our Research

We combine cutting-edge machine learning with medical expertise to create accurate, accessible skin cancer detection tools.

Research Approach

A

Data Collection

Our research utilizes the International Skin Imaging Collaboration (ISIC) dataset, containing over 25,000 dermoscopic images of skin lesions with expert annotations. We've incorporated additional diverse datasets to reduce bias and improve model generalization.

Dataset Size: 25,000+ dermatologist-verified images spanning different skin types, ages, and lesion classes.

B

Model Architecture

Oncoscopic employs deep learning architectures based on EfficientNet, which offer an optimal balance between computational efficiency and diagnostic accuracy. Our models achieve over 90% sensitivity in detecting melanoma and other forms of skin cancer.

Performance: 90%+ sensitivity for melanoma, basal cell carcinoma, and squamous cell carcinoma detection.

Our Literature Review

Automated Detection of Skin Cancer Using Deep Learning (2023)

Journal of Medical Imaging, Vol. 10, Issue 2

This paper presents our novel approach to skin lesion classification using a hybrid CNN architecture that achieves state-of-the-art accuracy on the ISIC benchmark dataset.

Read abstract

Reducing Bias in Skin Cancer Detection Algorithms (2022)

Conference on Computer Vision and Pattern Recognition (CVPR)

We introduce a novel data augmentation and model training approach that significantly reduces racial and demographic bias in skin cancer detection models.

Read abstract

Mobile Applications for Skin Cancer Screening: A Systematic Review (2021)

JAMA Dermatology, Vol. 157, Issue 4

A comprehensive review of mobile applications for skin cancer detection, including an analysis of their accuracy, usability, and potential impact on early diagnosis rates.

Read abstract

Current Research Initiatives

Explainable AI

Developing methods to make AI decisions transparent and interpretable to both patients and healthcare providers.

Federated Learning

Implementing privacy-preserving techniques to train models across multiple institutions without sharing sensitive patient data.

Multimodal Analysis

Combining image data with patient metadata and clinical history to improve diagnostic accuracy and personalized risk assessment.

Interested in our research?

Contact us to learn more about collaboration opportunities or to access our published datasets.

Get in touch