AI & Machine Learning
Skin Cancer Lesion Classifier
An Early-Stage CNN Model for Medical Image Analysis
The Problem
Automated analysis of dermatoscopic images is a complex computer vision challenge that could one day assist dermatologists in early-stage cancer detection. This requires robust data preprocessing and carefully designed neural network architectures.
The Solution
As a foundational exploration into medical imaging, I developed a Convolutional Neural Network (CNN) using PyTorch to classify different types of skin lesions from the HAM10000 dataset. The project focused heavily on the critical steps of data cleaning, image augmentation to prevent overfitting, and model training, achieving a validation accuracy of 70%. This was a valuable exercise in handling complex, imbalanced datasets and a crucial step in my machine learning journey.
Tech Stack
PythonPyTorchPandasScikit-learnMatplotlib
Status & Links
Completed
Key Features & Contributions
Complex Data Preprocessing
Implemented extensive image augmentation and normalization techniques to prepare the imbalanced dataset for training.
CNN Architecture
Designed and implemented a custom Convolutional Neural Network architecture in PyTorch tailored for image classification.
Model Training & Validation
Managed the complete training and validation pipeline, tracking metrics to evaluate model performance.
Foundation for Advanced Models
The experience gained from this project directly contributed to the higher accuracy achieved in the subsequent Plant Disease Detector project.