Using cystoscopy to segment bladder tumors with a multivariate approach in different color spaces

Area of Science:

• Medical Imaging
• Computational Pathology
• Urology

Background:

• Current bladder lesion diagnosis relies on subjective cystoscopy interpretation.
• Advanced techniques like virtual cystoscopy exist but do not utilize standard white light images.
• Automated analysis of traditional cystoscopy images can significantly improve early tumor detection and treatment.

Purpose of the Study:

• To develop a multivariate approach for segmenting bladder cystoscopy images.
• To automatically detect bladder tumors and aid physician diagnosis.
• To improve the efficiency of bladder tumor identification using white light cystoscopy.

Main Methods:

• A multivariate approach using Gaussian Mixture Models (GMM) to segment bladder cystoscopy images.
• Utilizing a Maximum a Posteriori (MAP) approach analyzing pixel intensities across RGB, HSV, and CIELab color spaces.
• Employing the Expectation-Maximization (EM) algorithm for optimal GMM parameter estimation.

Main Results:

• The proposed method achieves efficient two-class bladder tumor segmentation in RGB color space.
• The method demonstrates effectiveness even with poorly defined tumor shapes.
• Analysis indicated that excluding the L component from CIELab color space hinders tumor shape definition.

Conclusions:

• The developed image processing technique offers a more efficient way to segment bladder tumors from white light cystoscopy images.
• This automated approach has the potential to enhance early tumor detection and support clinical diagnosis.
• The study highlights the efficacy of the RGB color space for this segmentation task.