Publication: ระบบตรวจคัดกรองผู้ป่วยที่มีภาวะสมองบาดเจ็บไม่รุนแรงด้วยเทคโนโลยีปัญญาประดิษฐ์
2
0
Issued Date
2021
Resource Type
Language
tha
File Type
application/pdf
Access Rights
open access
Rights
ผลงานนี้เผยแพร่ภายใต้ สัญญาอนุญาตครีเอทีฟคอมมอนส์แบบ แสดงที่มา-ไม่ใช้เพื่อการค้า-ไม่ดัดแปลง 4.0 (CC BY-NC-ND 4.0)
Rights Holder(s)
มหาวิทยาลัยศรีนครินทรวิโรฒ
Suggested Citation
ธันยกานต์ ซื่อสัตย์สกุล, ศรัณยพงศ์ คุณากรพัฒนาการ (2021). ระบบตรวจคัดกรองผู้ป่วยที่มีภาวะสมองบาดเจ็บไม่รุนแรงด้วยเทคโนโลยีปัญญาประดิษฐ์. สืบค้นจาก: https://hdl.handle.net/20.500.14740/12012
Alternative Title(s)
Artificial intelligence based mild traumatic brain injury screening
Advisor(s)
Organization
Abstract
Traumatic brain injuries (TBI) caused by road traffic accidents have led to high rate of mortality and morbidity worldwide. Statistically, 70% to 90% of patients with mild TBI (mTBI) have been referred to primary care hospitals, where number of practitioners are limited for diagnosing a large number of cases. As a result, primary care hospitals are unable to diagnose patients and provide appropriate referrals to specialized hospitals. Therefore, this engineering project aims to study the pathology of mild traumatic brain injury from CT scans of the brain and investigate how to design an Al screening system for
mild traumatic brain injury patients to be precise and effective. We acquired CT scan dataset from 8 normal and 3 abnormal cases and converted the CT images to grayscale using Hounsfield units. Image enhancement methods were applied prior to feature extraction which includes histogram analysis, grey level co-occurrence matrices (GLCM), grey level run-length matrices (GLRLM) and fractal dimension (FD). All generated features were filtered by using Maximum Relevance and Minimum Redundancy (mRMR) method to select the top-ten features. To obtain the optimal feature set, we investigated the two feature selection methods based on the Sequential Forward Selection (SFS) and
Sequential Backward Elimination (SBE). The three classifiers: support vector machine (SVM), logistic regression (LR), and k-nearest neighbors (KNN) were employed to evaluate the performance of the binary outcome. The results showed that the processing workflow derived from mRMR, SBE, and KNN models produced the most effective Al system with the accuracy, precision, sensitivity, and specificity of 97.54%, 99.00%, 93.00%, and 99.60%, respectively.