Combined multiphoton microscopy and somatostatin receptor type 2 imaging of pancreatic neuroendocrine tumors

Abstract

Pancreatic neuroendocrine tumors (PNETs) are a rare but increasingly more prevalent cancer with heterogeneous clinical and pathological presentation. Surgery is the preferred treatment for all hormone-expressing PNETs and any PNET greater than 2 cm, but difficulties arise when tumors are multifocal, metastatic, or small in size due to lack of effective surgical localization. Existing techniques such as intraoperative ultrasound provide poor contrast and resolution, resulting in low sensitivity for such tumors. Somatostatin receptor type 2 (SSTR2) is commonly overexpressed in PNETs and presents an avenue for targeted tumor localization. SSTR2 is often used for pre-operative imaging and therapeutic treatment, with recent studies demonstrating that somatostatin receptor imaging (SRI) can be applied in radioguided surgery to aid in removal of metastatic lymph nodes and achieving negative surgical margins. However not all PNETs express SSTR2, indicating labeled SRI could benefit from using a supplemental label-free technique such as multiphoton microscopy (MPM), which has proven useful in improving the accuracy of diagnosing more common exocrine pancreatic cancers. Our work tests the suitability of combined SRI and MPM for localizing PNETs by imaging and comparing samples of PNETs and normal pancreatic tissue. Specimens were labeled with a novel SSTR2-targeted contrast agent and imaged using fluorescence microscopy, and subsequently imaged using MPM to collect four autofluorescent channels and second harmonic generation. Our results show that a combination of both SRI and MPM provides enhanced contrast and sensitivity for localizing diseased tissue, suggesting that this approach could be a valuable clinical tool for surgical localization and treatment of PNETs.

Automated classification of pancreatic neuroendocrine tumors using label-free multiphoton microscopy and deep learning

“…Our group has previously demonstrated the feasibility of supervised learning for the PNETs classification combined with label-free MPM. [2] However, such a classification approach requires handcrafted feature extraction procedures. Recently, convolutional neural networks (CNNs) have been highly successful in biomedical imaging tasks like image classification.…”