Table 1 Related work summary.

Souid et al.⁷

CNN used a Transfer learning strategy and wheighted loss to classify fourteen pathologies

Decent results in the pathology classification, achieving a 0.811 AUC score

Ait Nasser¹⁶

A similar approach to Souid et al.²¹ with some differentiation in the number of classes (from 8 to 14)

The method achieves very good results, the achieved AUC score is 0.949, however, the used dataset has a small number of samples which could limit the experimentation

Ayaz et al. ²²

Ensemble of CNNs : Inception, InceptionResNet, VGG16, VGG19, MobileNet

The method Achieved an AUC of 0.934

Cicero et al.²⁰

Using prived collected data from 2005 to 2015 consisting of 35,038 chest radiographs, the authors experimented with the Google LeNet5 CNN to construct an image classifier

The provided work has achieved an Aria Under Curve AUC of 0.964 for the classification of 5 pathologies and the beginning class. The score was obtained from testing over 2443 test samples

Rajaraman et al. ²³

The presented work provides a new strategy to target Tuberculosis pathology, where they used a bone suppression strategy over multiple datasets

The obtained results were significantly improved compared to no-bone suppression applied with ResNet CNN The ResNet-BN achieved an AUC of 0.963 while the ResNet achieved only 0.89