Research On Super-resolution Reconstruction Algorithm Of Pathological Images Based On Multi-scale Learnin

Pathological diagnosis is the gold standard for disease assessment in clinical practice.It is conducted in the form of microscopical-level inspection.Therefore,a very high-resolution pathological image that clearly describes the submicron-scale appearance is essential in the ear of digital pathology,which is not easily obtained.Recently,pathological image super-resolution(SR)has shown promising prospects in bridging this gap.However,the current research tendency is to improve reconstruction performance through complex network design while ignoring the full utilization of the multi-scale characteristics exhibited by pathological data at different levels.This paper proposes multi-scale learning methods to alleviate the problem.The main research work of this paper includes:Firstly,this paper propose a novel model that formulates the pathological image SR in a multi-task learning way.It adds an image magnification classification branch on top of the CNN-based SR network.Therefore,the learning objective is transformed into performing the SR while classifying the magnification as accurately as possible.The multi-task learning paradigm encourages the joint learning of multiscale mapping functions corresponding to multiple magnifications,and makes full use of multi-scale information between different magnification.It thus enables the learned model to adaptively accommodate the magnification variants,overcoming the problem that performing SR at different magnifications is treated as independent tasks in existing studies.Meanwhile,the incorporated classification label guides the model to learn a more powerful feature representation.In addition,GAN is introduced to improve the perception quality of the reconstructed image.Extensive experiments are conducted to validate the effectiveness of the model.It not only gains better performance in performing SR across magnifications and scaling factors,but also exhibits attractive plug-and-play nature on different CNN-based SR networks.Secondly,this paper propose a novel model that formulates the pathological image SR in a multi-path learning way.It accomplishes multi-scale learning by designing branch paths with different depths for information at multiple scales in an image,so as to make full use of the multi-scale information inside the single magnification image.Furthermore,the addition of the attention mechanism supports the adaptive detection of corresponding scale information by each path.Simultaneously,the model employs a simple and effective feature fusion technique to strike a balance between model performance and efficiency.Extensive experiments are conducted to validate that the model can better utilize and reconstruct image information thereby improving reconstruction accuracy and visual effects,and almost the same objective indicator test results are obtained with a model parameter quantity that is 5 times lower than the comparison network.