Assessing Bilateral Neurovascular Bundles Function with Pulsed Wave Doppler Ultrasound: Implications for Reducing Erectile Dysfunction Following Prostate Radiotherapy
Authors:
**g Wang,
Xiaofeng Yang,
Boran Zhou,
James Sohn,
Richard Qiu,
Pretesh Patel,
Ashesh B. Jani,
Tian Liu
Abstract:
This study aims to evaluate the functional status of bilateral neurovascular bundles (NVBs) using pulsed wave Doppler ultrasound in patients undergoing prostate radiotherapy (RT). Sixty-two patients (mean age: 66.1 +/- 7.2 years) underwent transrectal ultrasound scan using a conventional ultrasound scanner, a 7.5 MHz bi-plane probe and a mechanical stepper. The ultrasound protocol comprised 3 step…
▽ More
This study aims to evaluate the functional status of bilateral neurovascular bundles (NVBs) using pulsed wave Doppler ultrasound in patients undergoing prostate radiotherapy (RT). Sixty-two patients (mean age: 66.1 +/- 7.2 years) underwent transrectal ultrasound scan using a conventional ultrasound scanner, a 7.5 MHz bi-plane probe and a mechanical stepper. The ultrasound protocol comprised 3 steps: 1) 3D B-mode scans of the entire prostate, 2) localization of NVBs using color flow Doppler imaging, and 3) measurement of NVB function using pulsed wave Doppler. Five pulsed Doppler waveform features were extracted: peak systolic velocity (PSV), end-diastolic velocity (EDV), mean velocity (Vm), resistive index (RI), and pulsatile index (PI). In summary, this study presents a Doppler evaluation of NVBs in patients undergoing prostate RT. It highlights substantial differences in Doppler ultrasound waveform features between bilateral NVBs. The proposed ultrasound method may prove valuable as clinicians strive to deliver NVB-sparing RT to preserve sexual function effectively and enhance patients' overall well-being.
△ Less
Submitted 29 February, 2024;
originally announced March 2024.
Cross-Shaped Windows Transformer with Self-supervised Pretraining for Clinically Significant Prostate Cancer Detection in Bi-parametric MRI
Authors:
Yuheng Li,
Jacob Wynne,
**g Wang,
Richard L. J. Qiu,
Justin Roper,
Shaoyan Pan,
Ashesh B. Jani,
Tian Liu,
Pretesh R. Patel,
Hui Mao,
Xiaofeng Yang
Abstract:
Biparametric magnetic resonance imaging (bpMRI) has demonstrated promising results in prostate cancer (PCa) detection using convolutional neural networks (CNNs). Recently, transformers have achieved competitive performance compared to CNNs in computer vision. Large scale transformers need abundant annotated data for training, which are difficult to obtain in medical imaging. Self-supervised learni…
▽ More
Biparametric magnetic resonance imaging (bpMRI) has demonstrated promising results in prostate cancer (PCa) detection using convolutional neural networks (CNNs). Recently, transformers have achieved competitive performance compared to CNNs in computer vision. Large scale transformers need abundant annotated data for training, which are difficult to obtain in medical imaging. Self-supervised learning (SSL) utilizes unlabeled data to generate meaningful semantic representations without the need for costly annotations, enhancing model performance on tasks with limited labeled data. We introduce a novel end-to-end Cross-Shaped windows (CSwin) transformer UNet model, CSwin UNet, to detect clinically significant prostate cancer (csPCa) in prostate bi-parametric MR imaging (bpMRI) and demonstrate the effectiveness of our proposed self-supervised pre-training framework. Using a large prostate bpMRI dataset with 1500 patients, we first pretrain CSwin transformer using multi-task self-supervised learning to improve data-efficiency and network generalizability. We then finetune using lesion annotations to perform csPCa detection. Five-fold cross validation shows that self-supervised CSwin UNet achieves 0.888 AUC and 0.545 Average Precision (AP), significantly outperforming four comparable models (Swin UNETR, DynUNet, Attention UNet, UNet). Using a separate bpMRI dataset with 158 patients, we evaluate our method robustness to external hold-out data. Self-supervised CSwin UNet achieves 0.79 AUC and 0.45 AP, still outperforming all other comparable methods and demonstrating good generalization to external data.
△ Less
Submitted 17 March, 2024; v1 submitted 30 April, 2023;
originally announced May 2023.
