A multi-resolution textural approach to diagnostic neuropathology reporting

Area of Science:

• Neuropathology
• Medical Image Analysis
• Computational Pathology

Background:

• Accurate neuropathological diagnosis is critical for patient treatment.
• Distinguishing glioblastoma from metastatic cancer and assessing p53 protein status are key diagnostic challenges.
• Current methods can be subjective and time-consuming.

Purpose of the Study:

• To develop and evaluate a computer-aided diagnostic workflow for two critical neuropathology tasks.
• To enhance diagnostic accuracy and efficiency in intraoperative consultations and p53 status determination.
• To leverage texture analysis via discrete wavelet frames decomposition for improved feature extraction.

Main Methods:

• Texture analysis using discrete wavelet frames decomposition for feature extraction.
• Classification of glioblastoma versus metastatic cancer based on non-nuclei region textural features.
• Novel adaptive thresholding and a two-step classification approach for p53 staining intensity analysis (strong, moderate, weak, negative).

Main Results:

• Achieved up to 89.7% accuracy for glioblastoma and 87.5% for metastasis classification in intraoperative consultations.
• Demonstrated 85% average precision and 88% average sensitivity in detecting and distinguishing p53 cell types.
• Obtained 81% accuracy in classifying positive/negative p53 cells and 60% accuracy in sub-classifying positive cells by intensity.

Conclusions:

• The proposed computer-aided diagnostic workflow shows significant potential for improving accuracy in critical neuropathological diagnoses.
• Texture analysis is effective in differentiating glioblastoma from metastatic cancer.
• The p53 status classification method shows promise, achieving results comparable to neuropathologists.