Deep transfer learning - based automated detection of COVID-19 from lung CT scan slices

Abstract

In the proposed research work; the COVID-19 is detected using transfer learning from CT scan images decomposed to three-level using stationary wavelet. A three-phase detection model is proposed to improve the detection accuracy and the procedures are as follows; Phase1- data augmentation using stationary wavelets, Phase2- COVID-19 detection using pre-trained CNN model and Phase3- abnormality localization in CT scan images. This work has considered the well known pre-trained architectures, such as ResNet18, ResNet50, ResNet101, and SqueezeNet for the experimental evaluation. In this work, 70% of images are considered to train the network and 30% images are considered to validate the network. The performance of the considered architectures is evaluated by computing the common performance measures.<br><br>

Explainable Machine Learning for Early Assessment of COVID-19 Risk Prediction in Emergency Departments

“…In this way, the network is first trained on a similar task, such as lung cancer or pneumonia diagnosis. Next, the knowledge of the pre-trained network is “incremented” by applying a training phase where an augmented COVID-dataset is used [12] , [71] .…”

“…Importantly, considering that deep models have been highly criticized in the past for their “black-box” explanations, several deep models proposed for COVID-19 diagnosis [12] , [14] include a further interpretation step, applied to motivate the computed prediction. Among the various state-of-the-art methods for interpreting the predictions of deep models [72] , the mostly used are sensitivity analysis [12] , [73] , [74] , which allows the areas of highest activation to be identified, e.g., in the first hidden layer (since this layer is often considered as the one where base textural and color/gray level features are learned).…”

“…Importantly, considering that deep models have been highly criticized in the past for their “black-box” explanations, several deep models proposed for COVID-19 diagnosis [12] , [14] include a further interpretation step, applied to motivate the computed prediction. Among the various state-of-the-art methods for interpreting the predictions of deep models [72] , the mostly used are sensitivity analysis [12] , [73] , [74] , which allows the areas of highest activation to be identified, e.g., in the first hidden layer (since this layer is often considered as the one where base textural and color/gray level features are learned). Another common approach is output back-propagation as used by algorithms such as GRAD-CAM [13] , [75] or layer-wise relevance propagation [14] , [76] , which essentially back-propagate the activation in the output layer to understand which areas are the most relevant in the computation of the final decision.…”

“…Another interesting example of a deep learning model for CTs applies transfer learning to ResNet architectures and creates an augmented dataset by applying the usual image transformations to both the original image and the images obtained by wavelet decomposition. More precisely, instead of augmenting the dataset by transforming only the original image, wavelet decomposition is applied and also the images obtained from wavelet decomposition up to 3 levels are added to the training set [12] .…”

“…Given the promising results achieved by studies exploiting Artificial Intelligence (AI) and/or probabilistic models for outcome prediction [2] – [4] in bio-medical problems where human skill and know-how are not able to provide precise and reproducible solutions, in this worldwide health crisis, a great deal of research effort has been devoted to the development of Robotics and Data Analytics techniques [5] , [6] exploiting the potentials of AI methods to either predict, monitor, and combat the virus by simulating the virus spread or the time needed to recover [7] ensuring and promoting social distancing [5] , [8] , identifying early COVID-19 infections, or predicting patient outcome to improve patient care [8] – [11] . Thanks to such AI techniques, wearable devices and Web applications may be used by affected individuals for self-monitoring COVID-19 related symptoms, while clinicians are aided in the diagnosis of COVID-19 infection from either CT [12] , [13] or XRay lung images [11] , [14] , or in the prediction of patient mortality risk, progression to severe disease, intensive care unit admission, ventilation, intubation, or length of hospital stay [8] , [15] .…”