Automated Recognition of Cancer Tissues through Deep Learning Framework from the Photoacoustic Specimen

Area of Science:

• Biomedical Imaging
• Artificial Intelligence in Medicine
• Cancer Diagnostics

Background:

• Advancements in biomedical technology have expanded diagnostic capabilities.
• Multispectral photoacoustic imaging integrates optical and ultrasonic technologies for enhanced biomedical applications.
• Early cancer detection is crucial for reducing mortality rates.

Purpose of the Study:

• To develop and evaluate a hybrid optimization method for early cancer detection using multispectral photoacoustic imaging and deep learning.
• To improve the accuracy of cancer classification in ultrasound images.

Main Methods:

• A hybrid approach combining transfer learning-based cancer detection with multispectral photoacoustic imaging.
• Utilizing bilateral filtration (BF) for noise reduction in image processing.
• Employing lightweight LEDNet models for image segmentation and a feature extractor with particle swarm with ant colony optimization (PS-ACO) paradigm.
• Implementing a recurrent neural network (RNN) model for final image classification.

Main Results:

• The proposed PS-ACO-RNN technique effectively detects and classifies cancer in ultrasound images.
• The method demonstrates superior performance compared to existing approaches when validated on a benchmark database.

Conclusions:

• The PS-ACO-RNN method offers a promising advancement in early cancer detection through integrated imaging and deep learning.
• This hybrid approach shows significant potential for improving diagnostic accuracy in oncology.