This work contributes to the detection of regions of interest on images and their corresponding classification in medical imaging applications by introducing an image segmentation network that consists of four stages. In the first stage, multi-resolution processing is applied to outline regions where further segmentation and classification are to be conducted. Subsequently, a quad-tree division stage followed by a clustering stage deliver an image that is divided into unlabeled clusters. The output stage assigns each cluster to one class. This architecture offers flexibility in the input and output stages since this network can be fed with images of any size and the output stage can be implemented with any traditional classification model such as k-nearest neighbors, multi-layer perceptron, and support vector machine. Another contribution of this work is that this network does not rely on a very large number of annotated images for its training.