Automatic breast carcinoma detection in histopathological micrographs based on Single Shot Multibox Detector

Area of Science:

• Digital pathology
• Computational pathology
• Medical artificial intelligence

Background:

• Accurate histological classification by pathologists is crucial for effective breast cancer treatment and patient prognosis.
• Limited availability of pathologists necessitates AI-driven diagnostic support systems.
• AI can enhance the accuracy and efficiency of pathological diagnoses.

Purpose of the Study:

• To develop and evaluate an automatic breast lesion detection model using microscopic histopathological images.
• To assess the model's significance in assisting pathologists and medical students with diagnosis.
• To investigate the impact of AI assistance on diagnostic speed and accuracy.

Main Methods:

• A Single Shot Multibox Detector (SSD) model was developed for automatic breast lesion detection.
• The model was trained on 1361 microscopic images and validated on 315 images.
• Pathologists and medical students diagnosed images with and without AI model assistance.

Main Results:

• The SSD model achieved high diagnostic accuracies: 88.3% for 3-class (benign, non-invasive, invasive carcinoma) and 90.5% for 2-class (benign, malignant) classification.
• Mean intersection over union (IoU) was 0.59.
• Medical students' diagnostic accuracy significantly improved from 67.4% to 84.7% with AI assistance.
• AI assistance led to varied changes in diagnosis time, with some users experiencing reduced times.

Conclusions:

• An automated, high-speed breast lesion detection method using histopathological micrographs was developed.
• The AI system shows potential to support pathologists in laboratory histological diagnoses.
• AI-assisted diagnosis can improve accuracy and potentially efficiency in breast cancer pathology.