Figure 4
From: Weakly-supervised learning for lung carcinoma classification using deep learning
Examples of two carcinoma false positive prediction outputs using the fully-supervised method. Both were correctly predicted as non-neoplastic by the weakly-supervised method. (a) is a case of caseating granulomatous inflammation36 which consists of necrotic tissues (caseation necrosis is defined as a region in granulomas with eosinophilic, granular and cheese-like cellular debris with necrosis) and chronic inflammation. (b) is a case of granulomatous inflammation37, which is a distinctive form of chronic inflammation produced in response to various infection. The inflammatory tissue is the main cause of false positives due to its analogous tissue structure to cancer stroma.
