Optical coherence tomography combined with convolutional neural networks analysis can differentiate between tumor and healthy liver parenchyma ex vivo

Abstract

Purpose: Optical coherence tomography (OCT) is an imaging technology based on low-coherence interferometry which provides noninvasive, subsurface, high-resolution images of biological tissues. Potential clinical applications include intraoperative examination of resection margins or guidance of endoscopic biopsies, as a real-time adjunct to histological examination. In this study we investigated its ability to differentiate between healthy liver parenchyma and tumor ex vivo. Methods: Between June and August 2020, consecutive adult patients undergoing elective oncological liver surgery were included in this study. Fresh resection specimens were scanned ex vivo, before fixation in formalin, using a table-top OCT device at 1310nm wavelength. Scan dimensions were set at 3.0mm x 3.0mm x 2.5mm and 1024 x 1024 x 512 pixels, respectively. Scanned areas were excised and histologically examined separately. A pre-trained neural network (ResNet34) was used to match OCT-scans to their corresponding histological diagnoses, after splitting the dataset 70:30 into training and test data. The study was approved by the local ethics committee. Results: A total of 51 scans (comprising 50,176 images) were obtained from 19 patients, of which 18 (17,408 images) were of normal liver parenchyma and 33 (32,768 images) of tumor. The neural network analysis distinguished tumor from healthy liver parenchyma with a sensitivity and specificity of 92% and 87%, respectively. Conclusion: Optical coherence tomography combined with convolutional neural networks can distinguish between healthy and cancerous liver tissues with great accuracy ex vivo. Further studies are needed to improve upon these results and develop in vivo diagnostic technologies.