Classifying Microscopic Acute and Old Myocardial Infarction Using Convolutional Neural Networks

Area of Science:

• Forensic Pathology
• Computational Pathology
• Histopathology

Background:

• Convolutional Neural Networks (CNNs) have advanced significantly, impacting medical fields like radiology and histopathology.
• Research on CNNs in forensic/postmortem pathology is limited, primarily focusing on postmortem CT, unlike surgical histopathology.
• Myocardial infarction is a common postmortem finding requiring accurate identification and aging from histology.

Purpose of the Study:

• To investigate the efficacy of CNNs in identifying and aging myocardial infarction from histology slides.
• To compare the performance of commonly used CNNs in surgical histopathology for forensic applications.
• To establish a proof of concept for CNN application in forensic/postmortem histopathology.

Main Methods:

• Trained and tested four CNNs using 150 myocardial histology images (50 each: normal, acute myocardial infarction, old myocardial infarction).
• Utilized InceptionResNet v2 as one of the CNNs for classification tasks.
• Compared CNN performance in distinguishing normal myocardium from different stages of myocardial infarction.

Main Results:

• InceptionResNet v2 achieved over 95% accuracy in classifying normal myocardium from acute and old myocardial infarction.
• Demonstrated CNNs' capability to differentiate between normal and infarcted myocardial tissue.
• Indicated successful identification of myocardial infarction stages using automated analysis.

Conclusions:

• CNN technology shows significant potential as a screening tool in forensic/postmortem histopathology.
• CNNs can serve as computer-assisted diagnostic aids for identifying myocardial infarction in postmortem examinations.
• This study highlights a promising avenue for advancing forensic pathology through artificial intelligence.