A pioneering artificial intelligence tool to predict treatment outcomes in ovarian cancer via diagnostic laparoscopy.

Journal: Scientific reports
PMID:

Abstract

Ovarian cancer is associated with high rates of patient mortality and morbidity. Laparoscopic assessment of tumor localization can be used for treatment planning in newly diagnosed high-grade serous ovarian carcinoma (HGSOC). While spread to multiple intra-abdominal areas is correlated with worse outcomes, whether other morphological tumor differences are also associated with patient outcomes is unknown. Given the large volume of visual information in laparoscopic videos, we investigated whether deep-learning models can capture implicit features and predict treatment outcomes. We developed a novel deep-learning framework using pre-treatment laparoscopic images to assess clinical outcomes following upfront standard treatment, defined as short progression-free survival (PFS) (< 8 months) or long PFS (> 12 months). The deep-learning framework consisted of contrastive pre-training to capture morphological features of images and a location-aware transformer to predict patient-level treatment outcomes. We trained and extensively evaluated the model using cross-validation and analyzed the extracted features via UMAP visualizations and Grad-CAM saliency maps. The model reached an AUROC of 0.819 (± 0.119) on fivefold cross-validation and an out-of-fold AUROC of 0.807 on the whole dataset, successfully discriminating between patients with short PFS and long PFS using only laparoscopic images. Our approach demonstrates the potential of deep learning to simplify HGSOC triage and improve early treatment planning by accurately stratifying the patients based on minimally invasive laparoscopy at the diagnostic stage.

Authors

  • Xiaotian Ma
    School of Biomedical Informatics, The University of Texas Health Science Center at Houston, Houston, TX, USA.
  • Yu-Chun Hsu
    McWilliams School of Biomedical Informatics, The University of Texas Health Science Center at Houston, Houston, TX 77030, United States.
  • Amma Asare
    Department of Gynecologic Oncology and Reproductive Medicine, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
  • Kai Zhang
    Anhui Province Key Laboratory of Respiratory Tumor and Infectious Disease, First Affiliated Hospital of Bengbu Medical University, Bengbu, China.
  • Deanna Glassman
    Department of Gynecologic Oncology and Reproductive Medicine, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
  • Katelyn F Handley
    Department of Gynecologic Oncology, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL, USA.
  • Katherine Foster
    CHI Saint Joseph Medical Care, Lexington, KY, USA.
  • Khwahish Sharma
    Department of Gynecologic Oncology and Reproductive Medicine, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
  • Shannon Westin
    Department of Gynecologic Oncology and Reproductive Medicine, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
  • Amir Jazaeri
    Department of Gynecologic Oncology and Reproductive Medicine, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
  • Nicole D Fleming
    Department of Gynecologic Oncology and Reproductive Medicine, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
  • Pratip K Bhattacharya
    Department of Cancer Systems Imaging, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
  • Xiaoqian Jiang
    School of Biomedical Informatics, University of Texas Health, Science Center at Houston, Houston, TX, USA.
  • Anil K Sood
    The University of Texas MD Anderson Cancer Center, Houston, TX 77030, USA.
  • Shayan Shams
    School of Biomedical Informatics, The University of Texas Health Science Center at Houston, Houston, Texas, USA.