Deep Learning Techniques for Enhanced Classification of Gastrointestinal Images

Deep learning has revolutionized medical image analysis, particularly in the classification of gastrointestinal abnormalities. This paper proposes a classification model based on an ensemble learning approach, combining three Convolutional Neural Network (CNN) architectures: VGG16, DenseNet121, and InceptionV3. We first trained each model independently, then reused them with pre-trained weights from ImageNet to enhance performance. The outputs of the three models are concatenated to create a unified feature vector, which is then used for final classification through fully connected layers. This model was tested on the HyperKvasir dataset, a benchmark in gastrointestinal imaging. Our results demonstrate that using pretrained weights significantly improves accuracy compared to training from scratch, highlighting the effectiveness of the ensemble approach and pre-training in medical image classification.