Artificial intelligence (AI) is an emerging healthcare field poised to impact multiple aspects of cancer therapy. AI refers to the capability of machines to mimic humans. AI improves various aspects of cancer therapy, including workflow efficiency, diagnostic accuracy, proactive and predictive recommendations, precision medicine, image guidance, and drug discovery processes.1
Machine learning (ML) is a subset of AI that enables the machine to improve with experience by recognizing trends or patterns of data often containing predefined features. In contrast, deep learning is a subtype of ML that adapts and learns from large sets of unlabeled data.2
At the 2024 Cholangiocarcinoma (CCA) Foundation Annual Conference, 3 presentations discussed the application of AI learning– and ML-driven approaches to improve pathology, radiology, and drug discovery processes in CCA.
Julius Chapiro, MD, PhD, discussed emerging AI-driven radiology applications relevant to patients with CCA. In proof-of-concept studies, a deep learning algorithm trained to segment the liver and delineate hepatocellular carcinoma (HCC) on magnetic resonance imaging (MRI) automatically showed a high level of performance in automated detection and delineation of liver cancer and HCC on multiphasic contrast-enhanced MRIs.3
In another study, deep learning also demonstrated high performance in classifying common hepatic lesions with typical imaging features on multiphasic MRI. Dr Chapiro noted that milliseconds of processing time per lesion with deep learning could help make the clinical workflow more efficient.3
Additionally, the neural network outperformed radiologists in the classification accuracy of 60 hepatic lesions. An additional study used an ultrasound-based deep learning model to preoperatively differentiate HCC and intrahepatic CCA (iCCA) from HCC and CCA. Using MRI features, neural networks were also shown to predict treatment response to intra-arterial therapies for HCC and predict recurrence in early-stage HCC.3
Benjamin Goeppert, MS, presented an overview of the different applications of AI in improving pathological diagnosis in CCA trained with hematoxylin and eosin (H&E)-stained whole-slide images. Given that diagnosis of adenocarcinoma in the liver is a clinical challenge that significantly impacts clinical decision-making, the incorporation of AI-based algorithms has been shown to distinguish iCCA from liver metastasis using H&E-stained whole-slide images.
These approaches provide quick and reliable diagnosis while saving biopsy tissue samples for additional molecular analyses. AI-driven approaches have also enabled the prediction of iCCA subtypes, allowing the identification of actionable targets for therapeutic intervention.4
Mrinal Shekhar, PhD, discussed the advances in ML and molecular modeling in drug design. He noted that the conventional drug discovery process is expensive (>$2 billion), slow (~10 years), and inefficient. There are multiple failure points, with a high failure rate in early discovery and preclinical testing; only 1 to 2 candidates continue to clinical trials, underscoring the need for more robust identification and validation of drug targets.
Dr Shekhar noted that computer-aided drug design currently is a multistep collaborative process of “hit finding” (ie, identification of hits or candidate molecules binding to target), “hit-to-lead,” where experimentally validated hits are confirmed by physiochemical binding assays confirmation to become leads; and “lead optimization,” where experimentally validated leads are confirmed with cell-based/multitarget binding assays before they can be tested in clinical trials.
Dr Shekhar noted that active ML modeling helps accelerate virtual screening processes for candidate molecules and validation stages. Further ML modeling generates molecules while optimizing desired drug properties such as affinity, solubility, permeability, toxicity, and selectivity. Dr Shekhar also shared that computer simulation can help understand cancer resistance’s mechanistic basis.5
AI- and ML-driven approaches have the potential to revolutionize radiology, diagnostic pathology, and drug discovery processes, offering improved efficiency, accuracy, and speed. With additional research, these approaches are expected to eventually be integrated into the diagnosis and treatment of CCA to improve patient outcomes.
To sign up for our newsletter or print publications, please enter your contact information below.
