Article Text

Download PDFPDF
232 Sensitivity of a deep learning algorithm for detection of sentinel lymph node metastases in cervical cancer: a proof-of-concept study
  1. Ilse Gt Baeten,
  2. Nikolas Stathonikos,
  3. Jacob P Hoogendam,
  4. Cornelis G Gerestein,
  5. Geertruida N Jonges,
  6. Ronald P Zweemer and
  7. Paul JVan Diest
  1. UMC Utrecht, Utrecht, The Netherlands


Introduction/Background In early-stage cervical cancer, sentinel lymph node (SLN) mapping is a routine procedure for assessing nodal disease status. The current pathological ultrastaging protocol, involving serial sectioning and immunohistochemical staining, is essential for reliable SLN metastasis assessment but tedious, time-consuming, and expensive. Deep learning algorithms can aid the pathologist in adequately searching the serial sections for metastases, potentially reducing both workload and costs by improving accuracy and speed. This study evaluated the sensitivity of a deep learning algorithm for SLN metastasis detection in early-stage cervical cancer.

Methodology Whole slide images (WSI) of hematoxylin&eosin(H&E)-stained SLNs from early-stage cervical cancer patients diagnosed with SLN metastasis were retrospectively analysed. We employed a CE-IVD certified deep learning algorithm developed for detection of breast and colon cancer lymph node metastases (Metastasis Detection APP by Visiopharm). Sensitivity for off-label use in cervical cancer was assessed.

Results Twenty-one patients with early-stage cervical cancer were included: 15 patients with squamous cell carcinoma, five patients with adenocarcinoma, and one patient with clear cell carcinoma. Among these patients, 10 were diagnosed with macrometastasis and 11 with micrometastasis in at least one SLN. The algorithm was applied to evaluate H&E WSI of 47 SLN specimens, including 22 that were negative for metastasis, 13 with macrometastasis, and 12 with micrometastasis. The algorithm successfully detected all macro- and micrometastases, achieving a sensitivity of 100% for clinically relevant SLN metastasis detection. The algorithm generated a manageable number of annotations, which pathologists could rapidly review.

Conclusion This proof-of-concept study demonstrated high sensitivity of a deep learning algorithm for SLN metastasis detection in early-stage cervical squamous cell and adenocarcinoma, even though it was developed for breast- and colon cancer adenocarcinomas. Our findings highlight the potential of leveraging existing algorithms for off-label SLN metastasis detection in different cancer types. Prospective validation of this promising and clinically valuable tool is needed.

Disclosures RZ is proctor for robot-assisted surgery in gynaecological oncology on behalf of Intuitive Surgical Inc. PvD is member of the advisory board of Paige, VisioPharm, and Sectra. All other authors declare no conflicts of interest.

Abstract 232 Figure 1

Output of a deep learning algorithm applied to H&E WSI of sentinel lymph nodes in early-stage cervical squam adenocarcinoma with correct detection of macrometastasis (left) and micrometastasis (right). H&E, hematoxylin & eosin; WSI, whole slide images

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.