Skip to main content
Fig. 4 | BMC Medicine

Fig. 4

From: Deep learning radiomics based on contrast-enhanced ultrasound images for assisted diagnosis of pancreatic ductal adenocarcinoma and chronic pancreatitis

Fig. 4

Typical cases of our DLR model guiding radiologists to make correct decisions. The top panel shows two PDAC lesions. Most radiologists consider these lesions as CP lesions in the first reading, but 100% accuracy is achieved with access to the additional information generated from the DLR model. In these two cases, the score of the DLR model for PDAC is significantly higher than that of CP, and the area of the highlighted regions is large in the heatmaps. Most of them are distributed inside the tumor, which is consistent with the regular pattern of the PDAC lesions found. The bottom panel shows two CP lesions. Most radiologists consider these lesions as PDAC lesions in the first reading, and 100% accuracy is achieved with access to the additional information generated from the DLR model. In these two cases, the DLR model scores CP significantly higher than PDAC, and the area of the highlighted regions is small in the heatmaps and mostly distributed at the boundary of the ROI image, which is consistent with the regular pattern of CP lesions found. PDAC, pancreatic ductal adenocarcinoma; CP, chronic pancreatitis; ROI: region of interest; AI, artificial intelligence; DLR, deep learning radiomics

Back to article page