HYBRID EVENT: You can participate in person at London, UK or Virtually from your home or work.
ICC 2021

Anna Lian

Anna Lian, Speaker at Oncology Conferences
Brigham and Women’s Hospital/Harvard Medical School, United States
Title : Identification of Key Pathological Features Associated with Melanoma Survival by Artificial Intelligence NLP


Immunotherapy has reshaped the landscape of cancer treatment, showing promising treatment responses in melanoma. Currently, there are no pathological features which consistently predict treatment responses in melanoma; existing studies have involved small cohorts and shown conflicting results for all features except the depth of tumor invasion. Inflammatory regression and TILs are two pathological features that reflect host immune responses against cancer2-4. Currently, TILs are classified into three groups based on the presence of lymphocytes at the site of a tumor: absent, non-brisk and brisk5-6. However, there is no clear agreement as to the prognostic significance of these classifications7-10. In this study, we evaluate the prognostic value of host immune responses to melanoma by artificial intelligence (AI) approaches using natural language processing (NLP). In this retrospective cohort study that included over 2,500 patients with primary cutaneous melanoma were analyzed, NLP achieved an accuracy of 98% in feature extraction. Of the melanoma patients identified, 507 (19.3%) were deceased and 5-year survival was 74.3% (95% CI, 72.1% to 76.5%). Younger age, female sex, brisk TILs, lower Breslow thickness, lower mitotic rate, and absence of ulceration, microscopic satellites, or vascular/lymphatic invasion were independently associated with the increased survival probability (P <0.05). Absent, non-brisk, and brisk TILs were identified in 434 (16.5%), 1,916 (73.0%) and 274 (10.4%) patients, respectively. Compared to other status of TILs including absence and non-brisk TILs, brisk TILs had a 14.2% overall survival advantage at 5 years (adjusted HR, 0.7; 95% CI, 0.4 -1.0; P =.045). Our study indicate that the presence of brisk TILs is an independent prognostic factor for the overall survival of melanoma patients. This study also demonstrated that NLP can be a very effective and promising approach to build large longitudinal patient cohort that supports survival analyses.


Anna Lian has been working as trainee on natural language processing project to analyze pathology reports of cancer patients in Dr. Li Zhou’s laboratory at Brigham and Women’s Hospital.