Lesion Size Assessment on Multi-Frequency Single Transducer-Harmonic Motion Imaging Using Convolutional Neural Network

Medical imaging-derived tumor or lesion quantification provides clinically relevant information for diagnosis and treatment monitoring. Multi-frequency (multi-AM) single transducer harmonic motion imaging (ST-HMI) is an acoustic radiation force (ARF) based ultrasound elastography method that interrogates mechanical properties by transmitting a multi-frequency ARF excitation pulse and estimating the induced oscillatory displacements at 100-1000 Hz frequency simultaneously. In this study, an automated lesion size assessment method is presented using a CNN model, U2Net. The training-validation set contains 1094 multi-frequency ST-HMI images of inclusions with Young’s moduli from 6-70 kPa and diameters from 1.7-10.5 mm after data augmentation. For each of the 1094 inclusion images, ST-HMI-derived normalized peak-to-peak displacement (P2PD) images at 100-1000 Hz in the step of 100 Hz were used as the input of the network. Dice score, sensitivity, and specificity were used to evaluate the performance of the model in the test dataset containing ST-HMI imaging of inclusions, in vivo 4T1 breast cancer mouse tumors and ex vivo focused ultrasound (FUS)-induced lesions. The average (Dice score, sensitivity, specificity) were (0.924±0.041, 0.916±0.051, 0.991±0.006), and (0.871±0.068, 0.963±0.035, 0.946±0.038) in the inclusions, and mouse tumors, respectively. The absolute area estimation errors for FUS-induced lesions were 10% and 5%. These results demonstrated the potential of multi-frequency ST-HMI to determine lesion size in chemotherapy treatment monitoring and ablation therapy guidance using CNN.