Intraoperative Microscopic Assessment of Lymph Node Status in Gynecological Malignancies Using Full-Field Optical Coherence Tomography and Dynamic Cell Imaging: A Step From Image Guided Surgery Towards Deep Learning Applications
Intraoperative Microscopic Assessment of Lymph Node Status in Gynecological Malignancies Using Full-Field Optical Coherence Tomography and Dynamic Cell Imaging: A Step From Image Guided Surgery Towards Deep Learning Applications
The IMAGINE study is a monocentric, observational, prospective study evaluating the diagnostic performance of Full-Field Optical Coherence Tomography (FF-OCT) combined with Dynamic Cell Imaging (DCI) (Van Gogh System, AQUYRE Biosciences) for the intraoperative detection of lymph node metastases in fresh, unstained, ex vivo specimens from women undergoing surgery for gynecological malignancies (ovarian, endometrial, cervical, and vulvar cancer).
The primary objective is to assess the sensitivity, specificity, accuracy, negative predictive value, and positive predictive value of FF-OCT/DCI in detecting macrometastases (>2 mm), micrometastases (<2 mm), and isolated tumor cells (<0.2 mm), using hematoxylin-eosin histopathology as the reference standard, with results reported at both the lymph-node and patient level in accordance with STARD guidelines. Secondary objectives include determining the minimum detectable metastatic focus size, developing an FF-OCT/DCI imaging atlas of normal and metastatic lymph nodes, and developing a deep-learning algorithm for automated intraoperative assessment of lymph node status.
Based on an estimated 10% prevalence of nodal metastasis, a sample of 351 lymph nodes (approximately 160 patients) will be enrolled - prospectively at Fondazione Policlinico Universitario A. Gemelli IRCCS, and retrospectively from patients previously enrolled in the PROVE study (from February 2026) - over a planned study duration of 36 months.
Background and rationale. Accurate lymph node (LN) staging is a major determinant of prognosis and adjuvant treatment in gynecological malignancies. Reported rates of nodal involvement in apparently early-stage disease are approximately 14.2% in ovarian, 10% in endometrial, 15% in cervical, and 10% in vulvar cancer, rising to up to 55% in stage III-IV ovarian cancer. Systematic pelvic and/or para-aortic lymphadenectomy, performed to define nodal status, is associated with significant morbidity, including lymphocele in up to 38% of cases, wound infection/breakdown in 20-40% of vulvar cancer cases, and lymphedema in 13-48% of patients. Sentinel lymph node (SLN) biopsy has reduced this burden in early-stage disease, but intraoperative frozen section has limited sensitivity for micrometastases and isolated tumor cells (ITCs), is time-consuming, and is subject to mapping failure and "empty packet" phenomena (reported in up to 20% of obese patients with endometrial cancer). Preoperative imaging, including FDG-PET/CT, also has recognized false-negative rates (e.g., 12% for occult para-aortic involvement in locally advanced cervical cancer).
Full-Field Optical Coherence Tomography (FF-OCT) is a label-free, non-invasive microscopic imaging technology based on tissue reflectivity and light interference, generating real-time, high-resolution "en face" images of fresh, unstained tissue in under 10 minutes without fixation, sectioning, or staining. Dynamic Cell Imaging (DCI), performed on the same platform, adds quantifiable intracellular metabolic/contrast information complementary to FF-OCT morphology. In breast cancer, FF-OCT/DCI has previously shown sensitivity/specificity of 81.3%/90.3% (FF-OCT) and 91.7%/98.9% (DCI) for nodal metastasis detection (Yang et al., 2020). Unpublished preliminary data from the investigators (Pavone et al., submitted) report an accuracy of approximately 98% for FF-OCT versus 65% for frozen section in detecting metastatic foci >0.2 mm.
Hypothesis. FF-OCT combined with DCI can achieve higher diagnostic accuracy than intraoperative frozen section for detecting LN involvement in gynecological cancers, including detection of ITCs currently not reliably identifiable by frozen section, with potential to improve surgical decision-making (estimated 35% improvement, particularly in cervical cancer) and to reduce the "empty packet" rate (estimated 20% reduction), approaching 100% sample adequacy.
Study design and setting. This is a monocentric, observational, ambispective trial conducted at the Gynecologic Oncology Unit, Fondazione Policlinico Universitario A. Gemelli IRCCS, Rome, Italy. The prospective cohort includes patients referred for surgical LN harvest for gynecological malignancy; the retrospective cohort includes patients meeting the same eligibility criteria previously enrolled in the PROVE study (from February 2026 onward), for whom personal data processing complies with Article 110-bis of the Italian Privacy Code and the Sponsor's Data Protection Impact Assessment (DPIA), in accordance with Article 89 of Regulation (EU) 2016/679.
Procedures. Ex vivo LN specimens retrieved intraoperatively (or, for the retrospective cohort, previously collected/imaged specimens) are freed of fat and scanned with the CelTivity/Van Gogh biopsy system (AQUYRE Biosciences), which incorporates a Linnik interferometer with incoherent illumination. The device acquires "en face" images via a 1.24 mm × 1.24 mm scanning unit (~2 seconds per unit), with sequential scanning of adjacent units compiled into a broader field of view; native optical section thickness and resolution are 1 μm, allowing depth-resolved subsurface visualization at pre-established depths. Presence/size of macrometastases (>2 mm), micrometastases (<2 mm), and ITCs (<0.2 mm) are recorded from the optical images. Specimens are then paraffin-embedded with the scanned surface oriented to mirror the optical sample holder, sectioned, and stained with hematoxylin-eosin (H&E) for standard histopathological analysis, which serves as the reference standard. Optical and histologic images are manually co-registered for direct comparison. An attention-based deep-learning model is being developed by the Computational Pathology and Spatially-Integrated Omics (GSTeP) facility of Fondazione Policlinico Universitario A. Gemelli IRCCS to extract discriminative morphological/architectural features predictive of LN status, generating saliency maps to support interpretability of model predictions.
Sample size. The primary endpoint (sensitivity of FF-OCT for metastasis detection) is powered at the lymph-node level, assuming a 10% prevalence of nodal metastasis. A total of 351 nodes provides an estimated sensitivity of ~90% with a 95% CI half-width of 10%, and a precision of 4% for specificity. Assuming an average of 2-3 nodes sampled per patient, this corresponds to approximately 160 patients.
Statistical analysis. Categorical variables are summarized as counts/percentages; continuous variables as mean ± SD and median (IQR). Diagnostic accuracy metrics (sensitivity, specificity, accuracy, NPV, PPV) are reported with confidence intervals at both lymph-node and patient level, following STARD guidelines. Normality is assessed with the Kolmogorov-Smirnov test; the primary comparison (FF-OCT/DCI vs. frozen section/histology) is evaluated using paired parametric or non-parametric tests as appropriate.
Timeline. The prospective enrollment phase spans 24 months, with concurrent imaging data processing for radiomic/machine-learning model development, followed by an additional 12 months for in-depth analysis and model refinement (total planned duration: 36 months).
Inclusion Criteria:
Exclusion Criteria:
Presence of lymphoproliferative or granulomatous diseases affecting lymph nodes
matteopavone.21@gmail.com0630151