使用血管自動偵測演算法在脈衝式都普勒超音波

Abstract

在台灣，中風佔十大死因的第三位，中風造成的癱瘓也會造成社會經濟負擔。中風主要有分兩種，腦血管栓塞與腦溢血。目前都卜勒超音波系統，可用來探測心臟與顱內血管的流速，在心臟手術和中風病人間的用途已有廣泛之應用。但以經顱超音波的應用來說，系統的排線會限制操作的範圍與角度，另外手動式的血管訊號偵測非常花費時間。本研究目標便是以無線傳輸的方式，自動偵測病人腦中血管的情況。本研究發展了自動血管訊號偵測的演算法，並在仿體與人體上測試結果。此演算法能自動化及縮短尋找正確血管深度的時間，和加強最後血流分佈圖影像的清晰度。這些演算法技術先在單一和線性正列超音波探頭上設計。在先導研究中，使用5 MHz中心頻率探頭和CompuFlow1000組成的超音波系統，並用正列線性探頭來接收流速訊號對海藻膠仿體和頸動脈進行測試，來探討自動血管訊號偵測演算法的性能。在結果上，運用流速自動偵測演算法尋找到仿體血管的開始深度為34.18mm，和用波形自動偵測演算法在人體頸動脈上得到最佳的血管深度範圍為28.64~32.34mm。在影像後處理上，利用適應性脈衝藕合神經網絡(AD-PCNN)的原理達到接近SNR增值(8.70 to 21.72 dB)和壓抑雜訊 (10 dB)。至於無線傳輸的可行性也在國家晶片中心(NSOC)下和不同實驗室所研發出的無線經顱超音波系統 (wireless TCD)上測試。在一個蠕動幫浦和仿體血管架構，此超音波系統能正確的無線傳輸流速訊號。其高速的60 GHz 無線模組(~1 Gb/s) 和統一計算架構(CUDA)能提供快速的遠端資料傳輸與運算，不只能提供給病人和醫生觀察，也可傳出到醫院伺服器端做紀錄也提倡系統簡易操作性給醫療人員。未來我們希望能把AVDA和wireless TCD整合來達到臨床的自動化血管偵測。

Cerebral vascular diseases (strokes) account for the third highest cause of death in Taiwan. There are mainly two types of stroke, ischemic (clogging of brain vessel) and hemorrhagic (burst of brain vessel) strokes. The permanent disability caused by stroke also creates a huge cost for the society. Currently, Doppler ultrasound systems are used to detect vessel flows in the heart or intracranial region during the open-heart operation and stroke patients monitoring. However, the operation of transcranial Doppler (TCD) ultrasound systems requires experienced medical personnel and the Doppler gate seeking is done manually. Furthermore, the wires connecting the systems often restrict the angle and range of operation for the examiners. In order to perform faster diagnosis for stroke symptoms, automatic vessel detecting algorithm (AVDA) finding the Doppler gate (depth interval) of the phantom vessel and common carotid arteries is developed. The algorithm stresses on the depth interval where best flow signal can be obtained. In the early stage, a 5 MHz linear ultrasound transducer and CompuFlow 1000 are used to construct an ultrasound system. Under the phantom setup, the velocity scale AVDA locates the correct Doppler gate at starting depth of 34.18mm. Likewise, the vessel range for the Doppler gate is accurately determined to be at 28.64~32.34mm for the common carotid data from using the waveform AVDA. In the post image processing, adaptive pulse coupled neural network denoising (AD-PCNN) technique is used to improve the image SNR from 8.70 to 21.72 dB, and able to effectively inhibit additional noise at 10 dB. A wireless TCD system is designed under collaboration between various labs funded by National System on Chip’s (NSOC) grant. The wireless data transfer is inspected by experimenting over a phantom vessel with peristaltic pump. The wireless TCD system is able to correctly input and output flow signal from the transducer to the end display on the computer. The implemented 60 GHz wireless module and computed unified device architecture (CUDA) can provide fast data transferring (~1 Gb/s) and calculation of the Doppler spectrogram. Not only the spectrogram can be executed real-time for the patient diagnosis, the data can be transferred wirelessly into the hospital server. For future work, we wish to integrate the AVDA and the NSOC wireless TCD system to achieve automatic vessel detection in clinical settings.