A recent study published in the Journal of Cancer Research and Clinical Oncology reported optical coherence tomography (OCT) in combination with convolutional neural networks (CNN) can accurately differentiate between the intrahepatic cholangiocarcinoma (CCA) and normal liver parenchyma ex vivo with greater overall specificity, F1-scores, and sensitivity.

The researchers investigated cases of 11 consecutive adult patients who underwent elective liver resections for intrahepatic CCA between June 2020 and April 2021.

Read more about CCA therapies

The research team ex vivo scanned the areas of interest from the resection specimens before the formalin fixation through a table-top OCT device at a wavelength of 1310 nm.

In addition, the scanned areas were indicated and examined histologically, leading to a diagnosis for each scan. Moreover, the CNN was trained, validated, and then tested to match the OCT scans to their corresponding histological diagnoses with the help of a 5 × 5 stratified cross-validation process.

Study results indicated that the analysis included 24 3-dimensional scans (corresponding to about 85,600 individuals) taken from 10 patients. Furthermore, the 5 × 5 cross-validation results revealed that the model gained a mean F1 score of 0.94, while the sensitivity and specificity scores were 0.94 and 0.93, respectively.

“Our study demonstrated potential for incorporation of OCT and CNN in future in vivo clinical applications,” the authors highlighted. The incorporation of OCT and CNN in future vivo clinical trials can be achieved through various strategies, including the quick intraoperative examination of the liver resection margins resulting in decreased frozen sections and total operation time.

Another strategy could be the prompt and precise diagnosis of intrahepatic CCA during endoscopic retrograde cholangiopancreatography. To enable learning of more complex characteristics, the CNN architecture can be modified by increasing the number of hidden layers, processing the whole volume of data, or tuning hyperparameters.

Liver cancer, including hepatocellular carcinoma and intrahepatic CCA, is the 3rd deadliest and 6th most common cancer worldwide. Only a small percentage (10% to 40%) of patients with intrahepatic CCA are eligible for surgical therapy, and radical surgical removal is currently the only curative treatment for intrahepatic CCA without distant metastasis. There is an urgent need to improve diagnostic techniques to detect liver cancer earlier and accurately determine resection radicality during surgery.

OCT, a noninvasive imaging technology based on low-coherence interferometry, can potentially address these challenges. It produces high-resolution cross-sectional images with real-time imaging capabilities, making it an attractive imaging modality for liver cancer diagnosis.

Recent studies have shown that combining OCT with machine learning techniques, including CNN, can significantly improve diagnostic accuracy, the study team noted.

CNN can extract features from images, making it particularly useful for analyzing OCT images of liver cancer and differentiating between other types of cancer and normal tissue in other parts of the body. The combination of OCT with CNN or other machine learning techniques has the potential to revolutionize liver cancer diagnosis and provide significant clinical benefits for patients.


Wolff LI, Hachgenei E, Goßmann P. et al. Optical coherence tomography combined with convolutional neural networks can differentiate between intrahepatic cholangiocarcinoma and liver parenchyma ex vivo. J Cancer Res Clin Oncol Published online April 12, 2023. doi:10.1007/s00432-023-04742-x