Study On The Identification Of Atypical HCC Based On Diagnostic Semantics And Attention 3D-CNN Using CEUS

Hepatocellular Carcinoma(HCC)has high mortality.However,in medical images,it is difficult to visually distinguish atypical HCC from focal nodular hyperplasia(FNH).Most of the existing liver tumor classification methods are based on time intensity curve parameters,which rely on additional quantitative analysis software.What’s worst,they cannot be used for the identification of atypical HCC.In addition,for commonly used image features,due to the lack of diagnostic semantics,the clinical interpretability is poor,and doctors cannot understand the process of extracting information.These problems lead to the absence of automated diagnosis of atypical HCC.As for these problems,three innovative computer-aided diagnostic(CAD)models for FNH and atypical HCC are proposed in this paper.This paper firstly extracts three semantic features from the CEUS video to represent the clinical diagnostic semantics of tumors,and classifies them with support vector machine,making full use of the clinician’s prior knowledge.Then,a 3D-CNN network model based on the attention mechanism is proposed to extract the underlying features of the tumor and improve the generality of CAD model.Finally,based on the previous work,this paper proposed a CAD model that combined diagnostic semantics and 3D-CNN to ensure accurate classification and maximize the clinical interpretability of the model.In this paper,in order to evaluate their performance.342 cases of CEUS video were used to test the three models.The experimental results show that the three CAD models have achieved better classification performance than other similar types of CAD.To sum up,the models can provide clinicians with valuable diagnostic opinions.