Abstract:
Early detection of fatal skin cancer melanoma improves survival rates. Learning-based melanoma detection from dermoscopic images is promising. Since melanoma is rare, skin lesion databases have a high proportion of benign samples. Because the majority class dominates statistically, this imbalance biases classification models. We use dermoscopic image latent-space embedding for deep clustering to solve this problem. A novel center-oriented margin-free triplet loss (COM-Triplet) on convolutional neural network backbone image embeddings clusters. The proposed method forms maximally-separated cluster centers rather than minimizing classification error, making it less sensitive to class imbalance. We propose implementing COM-Triplet using Gaussian mixture model (GMM) pseudo-labels to avoid labeled data. Deep clustering with COM-Triplet loss outperforms triplet loss and competing classifiers in supervised and unsupervised settings.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
