Leonardo Tanzi

Nicolas Urbani, Sylvain Rousseau, Yves Grandvalet et al.

In many large-scale classification problems, classes are organized in a known hierarchy, typically represented as a tree expressing the inclusion of classes in superclasses. We introduce a loss for this type of supervised hierarchical classification. It utilizes the knowledge of the hierarchy to assign each example not only to a class but also to all encompassing superclasses. Applicable to any feedforward architecture with a softmax output layer, this loss is a proper scoring rule, in that its expectation is minimized by the true posterior class probabilities. This property allows us to simultaneously pursue consistent classification objectives between superclasses and fine-grained classes, and eliminates the need for a performance trade-off between different granularities. We conduct an experimental study on three reference benchmarks, in which we vary the size of the training sets to cover a diverse set of learning scenarios. Our approach does not entail any significant additional computational cost compared with the loss of cross-entropy. It improves accuracy and reduces the number of coarse errors, with predicted labels that are distant from ground-truth labels in the tree.

Leonardo Tanzi, Andrea Audisio, Giansalvo Cirrincione et al.

In recent years, the scientific community has focused on the development of CAD tools that could improve bone fractures' classification, mostly based on Convolutional Neural Network (CNN). However, the discerning accuracy of fractures' subtypes was far from optimal. This paper proposes a modified version of a very recent and powerful deep learning technique, the Vision Transformer (ViT), outperforming CNNs based approaches and consequently increasing specialists' diagnosis accuracy. 4207 manually annotated images were used and distributed, by following the AO/OTA classification, in different fracture types, the largest labeled dataset of proximal femur fractures used in literature. The ViT architecture was used and compared with a classic CNN and a multistage architecture composed of successive CNNs in cascade. To demonstrate the reliability of this approach, 1) the attention maps were used to visualize the most relevant areas of the images, 2) the performance of a generic CNN and ViT was compared through unsupervised learning techniques, and 3) 11 specialists were asked to evaluate and classify 150 proximal femur fractures' images with and without the help of the ViT, then results were compared for potential improvement. The ViT was able to correctly predict 83% of the test images. Precision, recall and F1-score were 0.77 (CI 0.64-0.90), 0.76 (CI 0.62-0.91) and 0.77 (CI 0.64-0.89), respectively. The average specialists' diagnostic improvement was 29% when supported by ViT's predictions, outperforming the algorithm alone. This paper showed the potential of Vision Transformers in bone fracture classification. For the first time, good results were obtained in sub-fractures classification, with the largest and richest dataset ever. Accordingly, the assisted diagnosis yielded the best results, proving once again the effectiveness of a coordinated work between neural networks and specialists.