Accurate assessment of liver fibrosis in the left liver lobe remains clinically challenging due to motion artifacts that compromise the reliability of shear wave elastography. This feasibility study introduces cardiovascular pulsing-based ultrasound strain imaging (CPUSI) integrated with deep learning, employing dual strain sequence strategies to assess its potential for detecting liver fibrosis in the left hepatic lobe by leveraging intrinsic cardiac motion. A total of 104 patients was enrolled for ultrasound image acquisition, which included B–mode imaging, acoustic radiation force impulse imaging (ARFI), and FibroScan measurements. The dataset was also used for CPUSI generation, extraction of proximal (cardiac-wall) and distal (intrahepatic) strain sequences, and calculation of strain metrics, including time-averaged strain (TAS) and the distal-to-proximal strain ratio (DPSR). Five deep learning models, namely recurrent neural network (RNN), long short-term memory (LSTM), gated recurrent unit (GRU), transformer, and temporal convolutional network (TCN), were trained using paired proximal and distal strain sequences to classify liver fibrosis stages, with histopathology serving as the reference standard. Diagnostic performance was evaluated using independent t-tests and area under the receiver operating characteristic curve (AUROC). CPUSI-derived TAS and DPSR significantly differentiated early-stage (F0–F1) from advanced-stage (F2–F4) fibrosis (p < 0.05). ARFI, FibroScan, CPUSI-derived strain metrics (TAS and DPSR), and the CPUSI-based deep learning framework using dual strain sequences achieved AUROC values of 0.83, 0.82, 0.72–0.73, and 0.95, respectively, with the highest performance observed for the LSTM model. The proposed CPUSI–deep learning framework offers a feasible noninvasive approach for left-lobe fibrosis assessment and may serve as a complementary tool to right-lobe-based elastography. Further studies with larger cohorts are warranted.
扫码关注我们
求助内容:
应助结果提醒方式:
