| 講演抄録/キーワード |
| 講演名 |
2023-05-18 15:30
Grad-CAM approach for Multiclass Magnetic Resonance Imaging Tumor detection and Classification ○Tahir Hussain・Shouno Hayaru(UEC) MI2023-4 |
| 抄録 |
(和) |
(まだ登録されていません) |
| (英) |
The growth of abnormal cells in the human brain causes brain tumors (BT). Early diagnosis becomes essential for timely treatment for patient survival. A radiologist examines magnetic resonance imaging (MRI) to diagnose and identify tumors through manual evaluation. This process is time-consuming and requires expertise for a complete understanding of tumor type and location. Existing methods suffer from unsatisfactory performance and lack of model explainability, especially in multiclass BT for clinical translation. However, physicians perceive the model results to be unsatisfactory due to Blackbox. Our study addresses these issues for multiclass classification of brain MRI tumor images and proposed a pre-train visual geometry group (VGG-19) that runs a new form of gradient-weighted class activation mapping (Grad-CAM) algorithm for model explainability. The Grad-CAM was used within the developed convolutional neural network (CNN) model, for the model explainability for BT diagnosis. The experimental findings show that the pre-train-VGG-19-Grad-CAM gives better classification and visualization results as compared to stat-off-art deep learning (DL) models with improved accuracy. The heatmap results can help radiologists to explain and validate the classification results by indicating the tumor region on the brain MRI and reducing misclassification. |
| キーワード |
(和) |
/ / / / / / / |
| (英) |
Grad-CAM / model explainability / pre-train-VGG-19 / / / / / |
| 文献情報 |
信学技報, vol. 123, no. 37, MI2023-4, pp. 10-13, 2023年5月. |
| 資料番号 |
MI2023-4 |
| 発行日 |
2023-05-11 (MI) |
| ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| PDFダウンロード |
MI2023-4 |