Smartphone-based Dermoscopy and AI-driven Diagnosis for Early Detection of Melanoma in Resource-Limited Settings

Project Abstract

Early detection of melanoma remains a critical challenge in resource-limited settings due to limited access to expert dermatologists and high-cost diagnostic tools. This study presents an integrated mobile dermoscopy platform coupled with AI-driven diagnostic models to enable accurate, rapid, and cost-effective melanoma screening in underserved populations. The system leverages a smartphone-based dermatoscope to capture high-resolution lesion images, which are preprocessed to normalize lighting, color, and focus variations typical of field conditions. A multi-stage AI pipeline combines both conventional machine learning features and deep learning representations to classify lesions on a risk scale, with a bias-reducing calibration step to address class imbalances common in dermatology datasets. We trained and validated the models on a curated dataset comprising diverse skin tones and lesion types, enriched with external datasets to improve generalizability across populations. To ensure practicality, the platform includes offline inference capabilities and lightweight on-device models, enabling operation without reliable internet access, while a secure cloud-based backend supports model updates, aggregated analytics, and teleconsultation workflows when connectivity is available. The study emphasizes data privacy and ethical considerations, implementing de-identification protocols and user consent workflows suited for community clinics. Performance metrics include sensitivity, specificity, area under the receiver operating characteristic curve (AUC-ROC), and negative/positive predictive values, with particular focus on maximizing sensitivity to minimize missed melanomas. Comparative analyses against conventional dermoscopic assessment and existing AI models demonstrate superior or comparable diagnostic accuracy with substantially reduced time-to-decision and lower operational costs. A user-centered evaluation involving primary care physicians and non-specialist health workers assesses usability, interpretability, and acceptance, informing iterative design improvements and training materials tailored to low-resource environments. The research also investigates the impact of image quality, device variability, and skin phototype on diagnostic performance, offering guidelines for standardizing image acquisition in field settings. Deployment scenarios explore integration with existing primary health care workflows, referral algorithms, and community screening campaigns, highlighting potential for scalability across rural clinics and mobile health units. The findings suggest that with proper calibration, robust preprocessing, and transparent AI explanations, smartphone-based dermoscopy can achieve clinically meaningful melanoma detection rates comparable to expert evaluation, while improving outreach and timeliness of care. The study concludes with recommendations for policy, capacity building, and future work toward fully autonomous triage tools and continuous learning systems that adapt to regional epidemiology and resource constraints.

Project Overview

What This Project Is About

---

The project explores using a smartphone to capture skin images and apply AI tools to help identify melanoma early. It focuses on making skin cancer screening affordable and accessible in places with limited medical resources.

---

---

The Problem It Addresses

---

Many people lack access to dermatology specialists and expensive imaging devices. Delays in detection can lead to worse outcomes. The project aims to fill this gap by pairing simple phone-based imaging with computer-aided analysis to flag suspicious lesions early.

---

---

Objectives of the Project

---

1. Assess how well a smartphone dermoscopy approach can capture useful skin images.
2. Develop or adapt an AI model to classify images as benign or suspicious.
3. Evaluate the tool's usability in resource-limited settings and identify user needs.
4. Compare AI-assisted results with expert assessments to gauge accuracy.
5. Propose guidelines for safe, ethical deployment in clinics or community health programs.

---

---

What You Will Do Step by Step

---

1. Review literature on mobile dermoscopy and AI in skin cancer detection.
2. Collect a small dataset of skin lesion images (with consent) using a smartphone adapter.
3. Preprocess images and train a simple AI classifier or adapt an existing one.
4. Validate the model against expert judgments and measure accuracy, sensitivity, and specificity.
5. Test the workflow with a non-specialist user and gather feedback on usability.

---

---

Expected Outcome

A practical, low-cost workflow that uses a smartphone and AI to help identify potential melanomas early, with preliminary evidence of accuracy and guidelines for safe use in limited-resource settings.