This study presents a workflow for rapid intraoperative cancer diagnosis that combines dynamic full-field optical imaging with deep learning analysis. During breast-conserving surgery, assessment of tumor margins commonly relies on frozen-section histopathology, a process that requires tissue destruction, specialized personnel, and approximately 30 minutes to deliver results. These constraints limit its routine intraoperative use and can delay surgical decision-making.
The proposed approach instead captures dynamic optical fluctuations from freshly excised, unstained tissue using dynamic full-field optical coherence tomography (D-FFOCT). These temporal signal variations arise from intracellular motion and microstructural organization, which differ systematically between malignant and benign tissue. By analyzing these dynamic patterns rather than static morphology, the method extracts functional tissue signatures that are informative for cancer diagnosis.
In a prospective study involving 129 patients undergoing breast surgery, the system was trained on 182 slides from 95 patients and evaluated on an independent test set of 42 slides from 34 patients. On this independent cohort, the model achieved 97.62% accuracy, 96.88% sensitivity, and 100% specificity, with only one invasive ductal carcinoma case misclassified. The complete imaging and analysis process required approximately 3 minutes, representing a tenfold reduction in diagnostic time compared with conventional frozen-section pathology.
The study also evaluated intraoperative margin assessment, a critical surgical task. In simulated margin analysis, 95.24% of margins (160 out of 168) were correctly classified, indicating the method’s potential to support real-time surgical decisions. Importantly, the imaging process is label-free and non-destructive, allowing all tissue specimens to remain intact for standard postoperative histopathology and downstream molecular testing.
Overall, this work provides strong clinical evidence that functional, dynamic optical imaging combined with AI analysis can deliver fast, accurate, and workflow-compatible intraoperative diagnostics. Rather than replacing pathology, the approach offers a practical decision-support tool that addresses time, resource, and tissue-preservation constraints inherent to current intraoperative assessment methods.
S. Zhang et al., “Potential rapid intraoperative cancer diagnosis using dynamic full-field optical coherence tomography and deep learning: A prospective cohort study in breast cancer patients,” Science Bulletin, vol. 69, pp. 1748–1756, 2024, doi: 10.1016/j.scib.2024.03.061.
