Abstract:
Advancements in artificial intelligence (AI) are revolutionizing various sectors and rapidly reshaping cancer research and personalized clinical care. The combination of big data and powerful computing capacity has unlocked the transformative potential of AI-based approaches, particularly deep learning and generative AI, in oncology. These innovations have significant applications in cancer diagnosis, treatment, and prognosis.
In this talk, I will present two studies that employ deep learning prediction models to detect cancer characteristics from digital pathology images. The first study focuses on predicting genomic mutations, tumor stage, grade, and survival outcomes directly from H&E-stained whole slide images (WSIs) of individual esophageal cancer patients by leveraging modern deep learning techniques. We achieved an average AUC of 0.90 for epithelium vs. stromal classification, and 0.87 for esophageal adenocarcinoma (EAD) vs. esophageal squamous cell carcinoma (ESCC) classification on the validation datasets. Additionally, these models predicted tumor stage, grade, and survival outcomes with an accuracy greater than 90%. TP53 gene alterations, occurring in over 50% of esophageal tumors, were predicted with an AUC of 0.91 from WSIs using our model.
The second study involved building a customized deep-learning model to detect YAP-positive drug-tolerant persister (DTP) cell states from whole histopathological image slides. The model achieved accuracies of 0.9091, 0.8949, and 0.902 in the training, validation, and testing datasets for lung cancer, respectively. With further clinical validation, this model could be implemented in routine cancer care to identify patient subpopulations with YAP1-activated tumors, who would benefit most from YAP1-targeted small molecule inhibitors.
These applications demonstrate the remarkable capabilities of AI in early detection, accurate diagnosis, optimization of cancer treatment protocols, and prediction of disease progression, recurrence, and patient survival. Furthermore, AI is also playing a key role in drug discovery, repurposing, and combination therapy strategies. We anticipate that the integration of AI technologies in cancer care will enhance the precision, efficiency, and personalization of patient management, ultimately improving clinical outcomes and quality of life for cancer patients.
Audience Take Away Notes:
- The general concept of artificial intelligence and its potentials in cancer research
- Application of deep learning algorithms on digital pathology images
- Practical examples of deep learning models in translational cancer research
