Abstract
PurposeFor the image classification problem, the construction of appropriate training data is important for improving the generalization ability of the classifier in particular when the size of the training data is small. We propose a method that quantitatively evaluates the typicality of a hematoxylin-and-eosin (H&E)-stained tissue slide from a set of immunohistochemical (IHC) stains and applies the typicality to instance selection for the construction of classifiers that predict the subtype of malignant lymphoma to improve the generalization ability.MethodsWe define the typicality of the H&E-stained tissue slides by the ratio of the probability density of the IHC staining patterns on low-dimensional embedded space. Employing a multiple-instance-learning-based convolutional neural network for the construction of the subtype classifier without the annotations indicating cancerous regions in whole slide images, we select the training data by referring to the evaluated typicality to improve the generalization ability. We demonstrate the effectiveness of the instance selection based on the proposed typicality in a three-class subtype classification of 262 malignant lymphoma cases.ResultsIn the experiment, we confirmed that the subtypes of typical instances could be predicted more accurately than those of atypical instances. Furthermore, it was confirmed that instance selection for the training data based on the proposed typicality improved the generalization ability of the classifier, wherein the classification accuracy was improved from 0.664 to 0.683 compared with the baseline method when the training data was constructed focusing on typical instances.ConclusionThe experimental results showed that the typicality of the H&E-stained tissue slides computed from IHC staining patterns is useful as a criterion for instance selection to enhance the generalization ability, and this typicality could be employed for instance selection under some practical limitations.
Highlights
Malignant lymphomas have more than 70 subtypes, and pathologists are required to identify the subtype from a set of microscopic images of a specimen that is invasively extracted from a patient to determine the treatment of the patient [1]
With the widespread use of whole slide images (WSIs) and the development of machine learning techniques, the image analysis of digital pathology has been accelerated, and there have been studies conducted on image classification [2,3,4], detection [5,6,7], segmentation [8,9,10], and survival prediction [11,12]
We demonstrate that instance selection based on the proposed typicality improves the generalization ability of the subtype classification model by a three-class classification experiment with 262 cases of malignant lymphoma
Summary
Our proposed method improves the generalization ability of the subtype classifier by selecting instances that have a typical appearance for each subtype and using them for training. SN ; D), where D denotes an N × N distance matrix, of which the (m, n) component is d(Sm, Sn), and fMDS denotes a function that maps L-dimensional vectors that represent the IHC staining patterns in the M-dimensional space based on the distances in order to make it easier to estimate probability density distribution of IHC stains. Typical instances would have the classifiable subtype-specific features in the H&E-stained WSIs because the pathologist could accurately infer the candidate subtype This quantitative measure can be employed as a typicality criterion for instance selection. If we know how typical each instance is, we can apply instance selection to a pathological image dataset to improve the generalization ability of the subtype classification, in which instances having the atypical appearance of H&E-stained WSIs are removed from the training data [13,14]. In Eq (5), L is the cross-entropy loss function for the bag class prediction
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: International Journal of Computer Assisted Radiology and Surgery
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
