Automatic cancer tissue detection using multispectral photoacoustic imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Pathology
• Cancer Diagnostics

Background:

• Traditional pathology for cancer detection in surgical specimens like those from thyroidectomy and prostatectomy is labor-intensive and costly.
• Microscopic examination of stained glass slides is the standard diagnostic method.

Purpose of the Study:

• To develop a fully automated deep learning algorithm for detecting malignant tissue in human thyroid and prostate specimens.
• To leverage multispectral photoacoustic (MPA) imaging for enhanced cancer region identification.

Main Methods:

• Utilized a convolutional neural network architecture for feature extraction from 3D MPA datasets.
• Employed a softmax function for simultaneous detection of thyroid and prostate cancer tissue.
• Trained the model on freshly excised ex vivo human tissue specimens.

Main Results:

• The automated deep learning model achieved a high predictive performance, with an area under the curve (AUC) of 0.96.
• Demonstrated successful differentiation of cancer regions from normal tissue based on MPA signal characteristics.

Conclusions:

• The developed deep learning model represents an advancement over previous machine and deep learning algorithms for cancer detection.
• This automated MPA imaging approach offers potential for immediate application in screening large numbers of surgical specimens.
• The model may serve as a foundation for future in vivo photoacoustic imaging analysis for cancer diagnosis.