Fast detection method for prostate cancer cells based on an integrated ResNet50 and YoloV5 framework

Area of Science:

• Medical imaging analysis
• Computational pathology
• Artificial intelligence in healthcare

Background:

• Prostate cancer screening relies on accurate identification of abnormal cells.
• Current methods can be time-consuming, impacting efficiency.
• Deep learning offers potential for automated and accelerated cell analysis.

Purpose of the Study:

• To develop a rapid deep learning-based detection method for prostate cancer abnormal cells.
• To enhance the speed and accuracy of precancerous screening.
• To promote the adoption of AI-assisted prostate cancer cell screening.

Main Methods:

• A two-stage deep learning approach was employed.
• Stage 1: ResNet50 for preliminary screening of cell clusters.
• Stage 2: YoloV5 for accurate localization and identification of abnormal cells.

Main Results:

• The proposed two-stage method was compared to a single-stage target detection model.
• Deep learning classification first identifies areas with potential abnormal cells.
• This reduces reasoning time by 50% with minimal accuracy loss.

Conclusions:

• A fast deep learning detection method for prostate cancer abnormal cells was successfully developed.
• The method significantly shortens reasoning time and enhances detection speed.
• This contributes to improved efficiency in prostate precancerous screening.