Classification of histological types in ovarian cancer using deep learning

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate histological typing of ovarian cancer is critical for predicting patient outcomes and treatment response.
• Certain ovarian cancer subtypes, like clear cell and mucinous carcinomas, present challenges due to chemotherapy resistance in advanced stages.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI) model for classifying histological types of ovarian cancer.
• To assess the performance of AI in distinguishing malignant from non-malignant ovarian tissues.

Main Methods:

• A deep learning model utilizing YOLOv8 for object detection and image classification was developed.
• The model was trained and validated on a dataset of 71 malignant and 400 non-malignant ovarian tissue images.
• Performance was evaluated using precision, recall, F1 scores, sensitivity, specificity, and accuracy on independent test datasets.

Main Results:

• The object detection model achieved Precision, Recall, and F1 scores of 0.502, 0.482, and 0.492, respectively, with a sensitivity of 66.8%.
• The classification model demonstrated a sensitivity of 53.0%, specificity of 86.5%, and accuracy of 79.2%.
• Five out of 200 non-malignant images were misclassified as malignant.

Conclusions:

• AI can classify ovarian cancer histology, but challenges exist due to morphological similarities between subtypes.
• Further research incorporating three-dimensional imaging is necessary to enhance AI model accuracy for ovarian cancer diagnosis.