Dense Pixel-Labeling For Reverse-Transfer And Diagnostic Learning On Lung Ultrasound For Covid-19 And Pneumonia Detection Article Swipe
YOU?
·
· 2021
· Open Access
·
· DOI: https://doi.org/10.1109/isbi48211.2021.9433826
· OA: W3165231405
We propose using a pre-trained segmentation model to perform diagnostic\nclassification in order to achieve better generalization and interpretability,\nterming the technique reverse-transfer learning. We present an architecture to\nconvert segmentation models to classification models. We compare and contrast\ndense vs sparse segmentation labeling and study its impact on diagnostic\nclassification. We compare the performance of U-Net trained with dense and\nsparse labels to segment A-lines, B-lines, and Pleural lines on a custom\ndataset of lung ultrasound scans from 4 patients. Our experiments show that\ndense labels help reduce false positive detection. We study the classification\ncapability of the dense and sparse trained U-Net and contrast it with a\nnon-pretrained U-Net, to detect and differentiate COVID-19 and Pneumonia on a\nlarge ultrasound dataset of about 40k curvilinear and linear probe images. Our\nsegmentation-based models perform better classification when using pretrained\nsegmentation weights, with the dense-label pretrained U-Net performing the\nbest.\n
