The scientific community has joined forces to mitigate the scope of the current COVID-19 pandemic. The early identification of the disease, as well as the evaluation of its evolution is a primary task for the timely application of medical protocols. The use of medical images of the chest provides valuable information to specialists. Specifically, chest X-ray images have been the focus of many investigations that apply artificial intelligence techniques for the automatic classification of this disease. The results achieved to date on the subject are promising. However, some results of these investigations contain errors that must be corrected to obtain appropriate models for clinical use. This research discusses some of the problems found in the current scientific literature on the application of artificial intelligence techniques in the automatic classification of COVID-19. It is evident that in most of the reviewed works an incorrect evaluation protocol is applied, which leads to overestimating the results.
Since the outbreak of the COVID-19 pandemic, computer vision researchers have been working on automatic identification of this disease using radiological images. The results achieved by automatic classification methods far exceed those of human specialists, with sensitivity as high as 100% being reported. However, prestigious radiology societies have stated that the use of this type of imaging alone is not recommended as a diagnostic method. According to some experts the patterns presented in these images are unspecific and subtle, overlapping with other viral pneumonias. This report seeks to evaluate the analysis the robustness and generalizability of different approaches using artificial intelligence, deep learning and computer vision to identify COVID-19 using chest X-rays images. We also seek to alert researchers and reviewers to the issue of “shortcut learning”. Recommendations are presented to identify whether COVID-19 automatic classification models are being affected by shortcut learning. Firstly, papers using explainable artificial intelligence methods are reviewed. The results of applying external validation sets are evaluated to determine the generalizability of these methods. Finally, studies that apply traditional computer vision methods to perform the same task are considered. It is evident that using the whole chest X-Ray image or the bounding box of the lungs, the image regions that contribute most to the classification appear outside of the lung region, something that is not likely possible. In addition, although the investigations that evaluated their models on data sets external to the training set, the effectiveness of these models decreased significantly, it may provide a more realistic representation as how the model will perform in the clinic. The results indicate that, so far, the existing models often involve shortcut learning, which makes their use less appropriate in the clinical setting.
Breast cancer is the most frequent in females. Mammography has proven to be the most effective method for the early detection of this type of cancer. Mammographic images are sometimes difficult to understand, due to the nature of the anomalies, the low contrast image and the composition of the mammary tissues, as well as various technological factors such as spatial resolution of the image or noise. Computer-aided diagnostic systems have been developed to increase the accuracy of mammographic examinations and be used by physicians as a second opinion in obtaining the final diagnosis, and thus reduce human errors. Convolutional neural networks are a current trend in computer vision tasks, due to the great performance they have achieved. The present investigation was based on this type of networks to classify into three classes, normal, benign and malignant tumour. Due to the fact that the miniMIAS database used has a low number of images, the transfer learning technique was applied to the Inception v3 pre-trained network. Two convolutional neural network architectures were implemented, obtaining in the architecture with three classes, 86.05% accuracy. On the other hand, in the architecture with two neural networks in series, an accuracy of 88.2% was reached.
The results showed an improvement of the contrast and edge visibility in phase-contrast images. These improvements may be important in the detection of small lesions and details.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.