کنفرانس مهندسی برق ایران

صفحه اصلی / سی و دومین کنفرانس بین المللی مهندسی برق

Enhancing Brain Tumor Classification in OCT Images using Local Phase Quantization Features

نویسندگان :

Naeem Eslamyeh Hamedani¹ Hasti Shabani²

1- دانشگاه شهید بهشتی 2- دانشگاه شهید بهشتی

کلمات کلیدی :

Brain tumor،local phase quantization،Optical coheence tomography،Structural features،Computer-aided diagnosis،K -means

چکیده :

Determining the boundary between healthy tissue and infiltrating tumors in brain tissue is considered one of the significant challenges for neurosurgeons. Optical coherence tomography (OCT) plays a vital role in assisting surgeons with brain tumor diagnosis. As a promising imaging modality, OCT offers fast and high-resolution imaging. However, interpreting OCT images is complex despite its advantages. Many studies have interpreted the information using intensity-based structural features. In this study, a proposed method aims to enhance diagnosis based on local phase quantization (LPQ) features. To the best of our knowledge, frequency-based features have not been utilized in any previous studies involving brain tumor classification with OCT images. Additionally, an automated K-means algorithm has been employed to quickly identify homogenous regions with high SNR in OCT images. The dataset comprises B-scans from 16 different patients, selected from regions with tumor infiltration exceeding 60% in the white matter region (WM>60%) and healthy tissues (WM0%). The results were achieved by utilizing SVM classification and employing adaptive synthetic sampling (ADASYN), coupled with grid-search cross-validation (Grid-SCV), to address data imbalance and optimize model parameters. The results, validated based on the accuracy (AC), sensitivity (SE), and specificity (SP), reached 98.53%, 98.14%, and 98.88%, respectively outperformed previous works and highlighted its potential as a computational diagnostic tool.