Diagnosis of Mesothelioma Using Image Segmentation and Class-Based Deep Feature Transformations.

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Mesothelioma is a rare, aggressive cancer often diagnosed late.
• Asbestos exposure is a primary cause.
• Limited data and complex imaging hinder timely diagnosis.

Purpose of the Study:

• To develop a novel hybrid model for accurate and timely mesothelioma diagnosis.
• To overcome challenges posed by limited datasets and complex tissue structures.
• To improve early detection rates for mesothelioma.

Main Methods:

• Integrated automatic image segmentation (SAM), transformer models (CaiT, PVT), and image transformation (Decoder, GAN, NeRV).
• Extracted class-specific features and transformed them into informative image representations.
• Utilized discriminative score, class centroid analysis, and SVM for final classification.

Main Results:

• Achieved 99.80% classification accuracy in mesothelioma diagnosis.
• Demonstrated effectiveness in handling limited data and complex tissue characteristics.
• Validated the model's capability for precise and efficient diagnosis.

Conclusions:

• The proposed hybrid model offers a highly accurate and efficient solution for mesothelioma diagnosis.
• Advanced techniques address key challenges in early and precise detection.
• The model shows significant potential for clinical application in oncology.