Efficient Lung Ultrasound Classification

Abstract

A machine learning method for classifying Lung UltraSound is here proposed to pro-
 vide a point of care tool for supporting a safe, fast and accurate diagnosis, that can also be useful during a pandemic like as SARS-CoV-2. Given the advantages (e.g. safety, rapidity, portability, cost-effectiveness) provided by the ultrasound technology over other methods (e.g. X-ray, computer tomography, magnetic resonance imaging), our method was validated on the largest LUS public dataset. Focusing on both accuracy and efficiency, our solution is based on an efficient adaptive ensembling of two EfficientNet-b0 models reaching 100% of accuracy, which, to our knowledge, outperforms the previous state-of-the-art. The complexity of this solution keeps the number of parameters in the same order as an EfficientNet-b0 by adopting specific design choices that are adaptive ensembling with a combination layer, ensembling performed on the deep features, minimal ensemble only two weak models. Moreover, a visual analysis of the saliency maps on sample images of all the classes of the dataset reveals where the focus is on an inaccurate weak model versus an accurate model.

Rep2former: a classification model enhanced via reparameterization and higher-order spatial interactions

“…In addition, prior research has provided insights for our design; for example, as shown in Ref. 41 suggests that increasing the network’s width and depth is an effective approach to improve model accuracy. Based on this notion, we have increased the network depth and replaced the convolution layer with a reparameterized multi-branch structure design.…”

Autoselection of the Ensemble of Convolutional Neural Networks with Second-Order Cone Programming

“…Recent studies in the literature Bruno et al (2022); Kolesnikov et al (2020); Sun et al (2017) leads to the conclusion that the use of pre-trained models and transfer learning increases the performance of the model and ensembling techniques outperform other state of the art models. For this reason, it is planned to test the SOCP model on an ensemble that will be created by using the models in the literature instead of the ensemble we randomly generated as a future work.…”

“…In one of the recent studies, ensemble technique is chosen to increase the complexity by training endto-end two EfficientNet-b0 models with bagging. Adaptive ensemble technique is used by fine-tuning within a trainable combination layer which outperforms different studies for widely known datasets such as CIFAR-10 and CIFAR-100 Bruno et al (2022). Since pre-trained models boosts efficiency while simplifying the hyperparameter tuning, increasing the performance on these datasets are achieved with the help of transfer learning and pre-training Kolesnikov et al (2020); Sun et al (2017).…”

Efficient Lung Ultrasound Classification

“…In this paper, we aim to introduce new methods and models for the analysis of LUS images in order to distinguish among SARS-CoV-2, pneumonia, and healthy conditions. The proposed new model, using EfficientNet-b0 [36] as a core, is based on a recent strategy for ensembling at the deep features level [37,38]. It reaches the state-of-the-art (SOTA) accuracy of 100% on a well-known public reference dataset.…”