Sperm hunting on optical microscope slides for forensic analysis with deep convolutional networks - a feasibility...

Area of Science:

• Forensic Science
• Computer Vision
• Biomedical Imaging

Background:

• Microscopic sperm detection is crucial in sexual assault investigations.
• Manual scanning of forensic samples is time-consuming, especially with low or absent sperm presence.
• Automated methods are needed to improve efficiency and accuracy in forensic analysis.

Purpose of the Study:

• To evaluate the efficacy of convolutional neural networks (CNNs) for automated sperm cell detection.
• To reduce the manual labor and time associated with microscopic examination of forensic samples.
• To develop a computational tool for rapid and reliable identification of spermatozoa.

Main Methods:

• Training a VGG19 network and a VGG19 variation using a dataset of 1942 images.
• Utilizing images containing both sperm cells and negative samples for robust model training.
• Applying the trained CNN models to locate sperm cells in microscopic images.

Main Results:

• Convolutional neural networks demonstrated capability in identifying sperm cells within microscopic images.
• The developed CNN models showed potential in reducing the extensive scanning time required for manual analysis.
• Successful localization of sperm cells was achieved, aiding in the validation of forensic samples.

Conclusions:

• CNNs offer a promising solution for automating microscopic sperm detection in forensic casework.
• The use of AI in forensic science can significantly enhance the efficiency of sample analysis.
• This automated approach can expedite the investigation process in sexual assault cases.