transfer learning radiomics based on multimodal ultrasound imaging for staging liver fibrosis supplementary

Objective: To propose a transfer learning (TL) radiomics model to efficiently combine information from grayscale and elastogram ultrasound images for accurate liver fibrosis staging.

METHODS: A total of 466 patients who underwent partial hepatectomy were included, including 401 with chronic hepatitis B and 65 without pathological fibrosis. All patients underwent elastography and liver stiffness measurement (LSM) 2-3 days before his surgery. We proposed a TL deep convolutional neural network to analyze grayscale modality (GM) and elastogram modality (EM) images.

The TL process was used for liver fibrosis classification with an Inception V3 network pretrained on ImageNet. We compared the diagnostic performance of TL and non-TL.

Single modality values, including GM and EM only, and multimodality values, including GM + LSM and GM + EM, were evaluated and compared with those of LSM and serological indices. Receiver operating characteristic curve analysis was performed to calculate the optimal area under the curve (AUC) for classifying fibrosis from S4, ≥S3, and ≥S2.

RESULTS: GM and EM TL showed higher diagnostic accuracy than non-TL, with significantly higher AUC (all p<0.01). Both single-modal GM and EM were superior to LSM and serum index (all p < 0.001). Multimodal GM + EM, GM + LSM, GM and EM alone, LSM, and biomarkers (all S < 0, 05).

CONCLUSIONS: Liver fibrosis can be staged with excellent performance by a transfer learning model based on a combination of grayscale and elastogram ultrasound images.

To learn more about fibrosis click below-

https://brainly.com/question/16389822

#SPJ4