Genetic algorithms and convolutional neural networks to aid the diagnosis of compression vertebral fractures

Vertebral Compression Fracture (VCF) is a vertebral body fracture related to compressive forces, with vertebral body partial collapse. VCFs may occur secondary to trauma, but non- traumatic VCFs may be secondary to osteoporosis fragility (benign VCFs) or tumors (malignant VCFs). In the case of non-traumatic VCFs, the investigation of etiology is usually necessary, since treatment and prognosis are dependent on the type of VCF. Currently, there has been great interest in using Convolutional Neural Networks (CNNs) for the classification of medical images because these networks allow the automatic extraction of interesting features for the classification in a given problem. However, CNNs usually require large databases that are often not available. Besides, these networks generally do not use additional information that may be important for classification. A different approach is to classify the image based on a large number of predefined features, an approach known as radiomics. In this work, we propose a hybrid method for classifying VCFs that uses features from three different sources: i) intermediate layers of CNNs; ii) radiomics; iii) additional information from the patients and image histogram. In the hybrid method proposed here, external features extracted from the images are inserted as additional inputs to the first dense layer of a CNN. A Genetic Algorithm (GA) was used to i) select a subset of visual, radiomic and clinical characteristics relevant to the classification of FVCs; ii) select hyper parameters that define the architecture of the proposed hybrid model for classification. Experiments using different approaches for the inputs indicate that combining information can be interesting to improve the performance of the classifier. (AU)