(For USM Staff/Student Only)

EngLib USM > Ω School of Electrical & Electronic Engineering >

Automated detection of embolic signal for stroke prevention

Automated detection of embolic signal for stroke prevention / Ng Yan Duan
Ultrabunyi Doppler Transcranial (TCD) merupakan alat untuk menentukan kewujudan embolisma untuk pesakit strok. Oleh sebab isyarat TCD mengandungi ciri-ciri istimewa semasa kehadiran embolus-embolus maka ini membolehkan pengesanan embolus dalam darah. Pada hari ini, pengawasan manusia digunakan dalam amalan klinikal tetapi kaedah tersebut memerlukan perhatian secara manual dan ketepatan akan merosot disebabkan faktor keletihan. Oleh itu, sistem pengesanan embolus secara automatik adalah diperlukan. Dalam kajian ini, isyarat TCD diambil daripada perisian TCDsimulator untuk proses pengumpulan data. Dalam projek ini, empat kaedah pengesanan telah dikaji. Kaedah pertama ialah keedah keamatan domain masa di mana nisbah keamatan embolus kepada darah akan digunakan. Seterusnya, kaedah keamatan domain frekuensi telah digunakan di mana spektrum frekuensi diperiksa untuk mencari komponen frekuensi yang bermagnitud tinggi. Kemudian, kaedah ketiga yang digunakan ialah kaedah hibrid keamatan masa-frekuensi yang merupakan gabungan antara kedua-dua kaedah di atas. Akhirnya. kaedah keempat adalah berasaskan alat pembelajaran mesin iaitu mesin penyokong vektor (SVM). Di sini, Pekali Kepstrum Frekuensi Mel (MFCC) digunakan sebagai ciri kepada pengelas. Prestasi experimen bagi setiap kaedah diukur melalui peratusan ketepatan dan kelajuan pemprosesan. Keputusan menunjukkan keempat-empat kaedah adalah berupaya untuk digunakan dalam aplikasi masa sebenar. Dalam ujian ini, pemerhatian manusia juga dibuat sebagai perbandingan ketepatan. Keputusan yang terbaik dicapai melalui kaedah hibrid keamatan masa-frekuensi. 90.74% (149/162) daripada isyarat berembolus dan 100% daripada isyarat tanpa embolus telah berjaya dikenalpastikan. Prestasi kaedah ini adalah menyakinkan kerana ketepatan yabg dicapai oleh pemerhatian manusia ialah 87.45% untuk isyarat berembolus dan 100% untuk isyarat tanpa embolus. Satu antaramuka pengguna bergrafik yang mesra pengguna juga telah dibinakan dalam kajian ini. _______________________________________________________________________________________________________ Transcranial Doppler (TCD) ultrasound is a tool to determine the occurrence of embolism in stroke patients. As the TCD signals have special features during the presence of emboli and thus allowing the detection of the emboli in blood. Currently, human observation is used in clinical practice, but it requires manual attention and the accuracy will deteriorate due to fatigue factor. Therefore, an automated emboli detection system is needed. In this study, the TCD signals are extracted from the TCDsimulator software for data collection process. In this project, four detection methods are investigated. The first method is time-domain intensity method where the emboli-to-blood intensity ratio was used in the detection. Subsequently, the frequency-domain intensity method is employed where the frequency spectrum is inspected to search for the frequency components with high magnitude. Then, the third method used, which is the time-frequency intensity hybrid method, is the combination of the two methods above. Finally, the fourth method is based on a machine learning tool, i.e. support vector machine (SVM). Here, Mel-Frequency Cepstral Coefficients (MFCC) is used as features to the classifier. The performance evaluations of each method are measured by accuracy percentage and processing speed. The result revealed that all methods are viable to be employed in real-time application. In this study, human observation is also done as the golden standard for accuracy comparison. The best result is achieved by the time-frequency intensity hybrid method. 90.74% (149/162) of the embolic signals and 100% of the non-embolic signals are successfully identified. The performance of this method is promising as the accuracy achieved by human observation is 87.45% for embolic signals and 100% for non-embolic signals. A user-friendly graphical user interface is also developed in this study.
Contributor(s):
Ng Yan Duan - Author
Primary Item Type:
Final Year Project
Identifiers:
Accession Number : 875007978
Language:
English
Subject Keywords:
(TCD); ultrasound; emboli
First presented to the public:
6/1/2015
Original Publication Date:
3/20/2019
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Electrical & Electronic Engineering
Citation:
Extents:
Number of Pages - 93
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2019-03-20 16:54:51.446
Submitter:
Mohd Jasnizam Mohd Salleh

All Versions

Thumbnail Name Version Created Date
Automated detection of embolic signal for stroke prevention1 2019-03-20 16:54:51.446