Vascular graph network for ovarian lesion classification using optical-resolution photoacoustic microscopy.

Area of Science:

• Biomedical optics
• Medical imaging
• Computational pathology

Background:

• Ovarian lesion diagnosis is complex due to varied presentations.
• Benign conditions like endometriosis can mimic ovarian cancer.
• Ovarian vasculature changes significantly in cancerous versus benign states.

Purpose of the Study:

• To investigate differences in ovarian vasculature between cancer and benign conditions using OR-PAM.
• To develop a computational model for classifying ovarian lesions based on vascular features.
• To assess the diagnostic performance of the developed model.

Main Methods:

• Optical-resolution photoacoustic microscopy (OR-PAM) was used to acquire vascular data from 94 ovarian specimens.
• OR-PAM data were transformed into vascular graphs with physical properties.
• A custom vascular graph network (VGN) was developed for classification.

Main Results:

• VGN demonstrated stable predictions even with small sample areas (3 mm × 0.12 mm).
• VGN achieved 79.5% accuracy and an AUC of 0.877 for cancer diagnosis.
• Overall five-class classification accuracy for VGN reached 73.4%.

Conclusions:

• VGN effectively leverages vascular network topology for ovarian lesion classification.
• OR-PAM combined with VGN shows promise for improving ovarian cancer diagnosis.
• This approach offers a potential tool to distinguish malignant from benign ovarian pathologies.