研究目的
To develop algorithms for Computer-Aided Diagnosis (CAD) for brain tumors from T2 MRI images using texture analysis and classification methods, and to prove that unbiased input data for training requires a specific ratio of healthy to non-healthy tissue MRIs.
研究成果
The ANN classifier with GLCM features achieved the best sensitivity of 100% with unbiased input data (35% healthy, 65% non-healthy ratio). GLCM texture analysis is more efficient than DWT for this application. The input data ratio is critical for unbiased AI training, and further research with larger datasets is recommended.
研究不足
The study uses a relatively small dataset, which may limit generalization. The ANFIS classifier requires a large amount of data, leading to potential inaccuracies with the available dataset. The methods are applied only to T2 MRI images, and results may vary with other MRI sequences or larger datasets.
1：Experimental Design and Method Selection:
The study combines texture analysis methods (GLCM and DWT) with classification methods (ANN and ANFIS) for CAD of brain tumors from T2 MRI images without pre- or post-processing.
2：Sample Selection and Data Sources:
The database includes 202 non-healthy MRIs from MICCAI BraTS 2015 and Harvard databases, and 18 healthy MRIs from a Greek hospital, segmented by a neurosurgeon. Subsets are created with 35% healthy and 65% non-healthy ratios for unbiased training.
3：List of Experimental Equipment and Materials:
MRI images in various formats (mha, gif) converted to jpeg, with skull stripping and sharpening filters applied.
4：Experimental Procedures and Operational Workflow:
Skull stripping is performed using thresholding and morphological operations, followed by high-pass Gaussian filtering for sharpening. Texture features are extracted using GLCM (13 features) and DWT with PCA for dimensionality reduction. Classification is done using ANN (with scaled conjugate gradient or Levenberg-Marquardt backpropagation) and ANFIS.
5：Data Analysis Methods:
Performance is evaluated using sensitivity, specificity, and accuracy metrics from confusion matrices.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
