Role of Metrics in Medical Image Analysis based on Unsupervised Machine Learning

Abstract

Background and Purpose: The paper point out the role of metrics and these mathematical properties in a resolution of problem by a machine learning. A methodology of building an unsupervised machine learning is described, following by an application of the model to a medical image analysis and an automatic diagnosis. Our model is based mainly on clustering.

Methods: Considering the objectives of the medical image analysis such as to detect patterns on the image and to guide the diagnosis, we focus our interests on the unsupervised machine learning, specifically the clustering approach based the centroid and density models. These two models allow the data analysis and classification. Furthermore, we considered that classification can be done by attribution criteria such as object semantic and similarity criteria.

Results: We have demonstrated the application of unsupervised learning for medical image analysis and diagnosis. Our model was tested on 40 different images of samples. The accuracy in our unsupervised machine learning means detection and correct classification of all necessary objects in a single image.

Conclusions: Asking the machine to make a suitable grouping of the objects of an image in classes without human intervention by algorithms, this is the goal of unsupervised machine learning. To achieve this challenge, metrics play an important role. We built the formal model based on a set of rules and functions that can analyse and classifies objects’ image in classes. Our model as the benefit to be used for semi-supervised machine learning co-clustering applications. This can make it easier to label a large volume of data.