OC01.04: *AI‐driven ultrasound detection of ovarian cancer that generalises: an international multicentre validation study
Year: 2024
Type: conference-abstract
Abstract: A critical shortage of expert ultrasound examiners has raised concerns of unnecessary interventions and delayed diagnoses. Artificial intelligence (AI)-driven diagnostic support has the potential to alleviate this burden and improve patient outcomes. Deep learning has shown promising results in ovarian cancer detection; however, external validation is lacking. Our aim was to develop a deep learning model for ovarian cancer detection and evaluate its robustness and ability to generalise in a large multicentre setting. We retrospectively collected 17,119 ultrasound images from 3,652 women with an ovarian lesion (2,224 benign, 1,428 malignant) from 20 centres in eight countries. A total of 2,718 cases were reviewed by a minimum of 7 expert and 6 non-expert examiners. For each centre in turn, we trained a transformer-based model using data from the remaining centres and compared the models' performance with that of human examiners. Furthermore, we simulated and assessed how these models could be used in AI-assisted triage and compared it to current practice. The models demonstrated robust performance across centres, ultrasound systems, and histological diagnoses, with an area under the receiver operating characteristic curve of 0.93 (95% CI 0.92–0.94) and F1 score of 83.7% (95% CI 82.3–85.0) on cases from unseen centres. They outperformed both expert and non-expert examiners, with F1 scores of 79.6% (95% CI 78.3–81.3) and 74.7% (95% CI 73.0–76.3), respectively. The models were further shown to produce well-calibrated predictions. In a retrospective simulation, AI-assisted diagnostic support reduced the number of referrals to experts by 63% compared to current practice, while increasing the diagnostic performance (F1 83.0% vs 77.5%). The models exhibit strong generalization and outperform expert examiners. Introducing AI-driven diagnostic support into the clinical workflow may reduce human resource demands, while improving diagnostic performance.
DOI: 10.1002/uog.27724
Authors: F. Christiansen, E. Konuk, A. Ganeshan, R. Welch, J. Palés Huix, A. Czekierdowski, F. Leone, L.A. Haak, R. Fruscio, A. Gaurilcikas, D. Franchi, D. Fischerová, E. Mor, L. Savelli, M. Pascual, M.J. Kudla, S. Guerriero, F. Buonomo, K. Liuba, N. Montik, J. Alcazar, E. Domali, N.P. Pangilinan, C. Carella, M. Munaretto, P. Šašková, D. Verri, C. Visenzi, P. Herman, K. Smith, E. Epstein
Institutions: Karolinska Institutet, KTH Royal Institute of Technology, Science for Life Laboratory, Medical University of Lublin, Luigi Sacco Hospital, Ústav pro Péči o Matku a Dítě, European Institute of Oncology, Charles University, Fondazione Poliambulanza Istituto Ospedaliero, University of Bologna, Hospital Universitario Dexeus, Medical University of Silesia, University of Cagliari, IRCCS Materno Infantile Burlo Garofolo, Skåne University Hospital, Marche Polytechnic University, Clinica Universidad de Navarra, Medical Center Manila, Ospedale G.B. Morgagni - L.Pierantoni
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4402358336