ENHANCEMENT OF THE AUCMEDI FRAMEWORK FOR MEDICAL IMAGE CLASSIFICATION

• Main Author: Gilang Pramudya, Aloysius
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85082
• Institution: Institut Teknologi Bandung

This thesis discusses the enhancement of the AUCMEDI framework for classifying medical problems from medical images. The primary challenge addressed is the limitation in the variety of image pre-processing methods, data augmentation techniques, and evaluation metrics available in the AUCMEDI framework. To overcome these limitations, the enhancement includes the addition of image pre- processing methods such as contrast enhancement using Contrast Limited Adaptive Histogram Equalization (CLAHE) and noise reduction with Box, Median, and Gaussian filters. Furthermore, data augmentation through Generative Adversarial Networks (GAN) is incorporated to enrich the diversity of the training data. New evaluation metrics, such as Positive Likelihood Ratio (LR+) and Brier Score, are also implemented to provide a more comprehensive assessment of model performance. The methodology applied in this enhancement involves domain engineering to identify specific requirements in the medical image classification domain and a hot spot-driven framework development approach to enhance flexibility. Testing results indicate that this enhancement successfully improves the flexibility and functionality of AUCMEDI in constructing classification pipelines for medical images across various dataset characteristics.