Related Experiment Video
Updated: Oct 10, 2025
08:49
A 3D Organotypic Melanoma Spheroid Skin Model
Published on: May 18, 2018
16.0K
Unsupervised Phenotype-Based Clustering of Clinicopathologic Features in Cutaneous Melanoma
Sarem Rashid1, Nikolai Klebanov2, William M Lin2
1Wellman Center for Photomedicine, Massachusetts General Hospital, Boston, Massachusetts, USA.
JID Innovations : Skin Science From Molecules to Population Health
|December 15, 2021
Summary
Machine learning identified distinct cutaneous melanoma subtypes. De novo melanomas, lacking precursor lesions, showed more aggressive features than nevus-associated melanomas.
Area of Science:
- Oncology
- Dermatology
- Computational Biology
Background:
- Current cutaneous melanoma classifications lack concordance, being limited to morphology or molecular data.
- A need exists for integrated clinicopathologic classifications to improve melanoma subtyping.
Purpose of the Study:
- To apply unsupervised machine learning (k-medoids clustering) to clinicopathologic data for novel melanoma subset identification.
- To investigate differences between computer-defined melanoma clusters using mixed variables from surgical pathology reports.
Main Methods:
- Unsupervised k-medoids clustering applied to 2,978 primary cutaneous melanomas.
- Analysis of mixed clinicopathologic variables including morphology, staging, and growth phase.
Main Results:
- Five distinct melanoma clusters were identified, falling into two subspaces: nevus-associated and de novo.
- De novo melanomas exhibited increased mitogenicity, ulceration, thickness, and distinct growth phase characteristics compared to nevus-associated melanomas.
Conclusions:
- Machine learning-driven clinicopathologic clustering offers a biologically relevant approach to melanoma classification.
- This method facilitates the discovery of new melanoma subtypes and potentially novel genomic associations.
