Skip to content

Team · Team Project | Computer Vision Capstone & Research

Skin Cancer Prediction and Detection

Connected to ICMCTT 2024 publication · 4th author

A computer-vision decision-support system leveraging transfer learning and image preprocessing to assist in early skin-lesion assessment — framed as clinical assistance, not autonomous diagnosis.

My contribution

As a team member I worked on dataset preparation, model experimentation, evaluation loops, and integrating the prediction pipeline into a usable workflow. I kept the narrative honest: AI as screening support under expert oversight.

Python · OpenCV · CNN · VGG16 · ResNet50 · Transfer learning · Isolation Forest · One-Class SVM · Model evaluation

Problem

Early visual screening of dermatological lesions suffers from high inter-observer variability and subtle morphological differences between benign and malignant cases.

Solution & architecture

Built an image-classification pipeline combining OpenCV preprocessing (Gaussian blurring, histogram equalisation) with deep convolutional architectures (CNN, VGG16, ResNet50 via transfer learning) and anomaly-detection baselines (Isolation Forest, One-Class SVM).

Modeling & evaluation

Comparative benchmarking across fine-tuned CNN backbones with precision/recall tracking to minimise false negatives. Outputs were designed as decision support a reviewer could interpret — not a diagnostic replacement.

My contribution

Team contribution across data prep, training/evaluation iterations, and workflow integration, with explicit attention to scope boundaries and responsible framing.

Impact & recognition

Connected to peer-reviewed research — "Earlier Cancer and Tumor Detection Using Deep Learning Techniques" (ICMCTT 2024, 4th author) — demonstrating strong classification performance in related oncology imaging workflows. Portfolio claims stay tied to that publication record rather than invented clinical metrics.