Setting the Research Agenda for Clinical Artificial Intelligence in Pancreatic Adenocarcinoma Imaging

Abstract

Pancreatic ductal adenocarcinoma (PDAC), estimated to become the second leading cause of cancer deaths in western societies by 2030, was flagged as a neglected cancer by the European Commission and the United States Congress. Due to lack of investment in research and development, combined with a complex and aggressive tumour biology, PDAC overall survival has not significantly improved the past decades. Cross-sectional imaging and histopathology play a crucial role throughout the patient pathway. However, current clinical guidelines for diagnostic workup, patient stratification, treatment response assessment, and follow-up are non-uniform and lack evidence-based consensus. Artificial Intelligence (AI) can leverage multimodal data to improve patient outcomes, but PDAC AI research is too scattered and lacking in quality to be incorporated into clinical workflows. This review describes the patient pathway and derives touchpoints for image-based AI research in collaboration with a multi-disciplinary, multi-institutional expert panel. The literature exploring AI to address these touchpoints is thoroughly retrieved and analysed to identify the existing trends and knowledge gaps. The results show absence of multi-institutional, well-curated datasets, an essential building block for robust AI applications. Furthermore, most research is unimodal, does not use state-of-the-art AI techniques, and lacks reliable ground truth. Based on this, the future research agenda for clinically relevant, image-driven AI in PDAC is proposed.

Artificial Intelligence in Pancreatic Ductal Adenocarcinoma Imaging: A Commentary on Potential Future Applications

“…Current research separates detection, defined as the distinction between PDAC patients and healthy control subjects, from differential diagnosis, defined as the distinction between PDAC and other types of pancreatic lesions. 13 The previously described studies indicate that AI trained with large data sets can approach expertlevel performance. 22,23 However, both studies focus on binary classification as opposed to differential diagnosis, and the evidence for radiologists' performance is limited because no multiinstitutional reader studies have been conducted.…”

“…11 However, AI research in PDAC is still at a preliminary stage compared with other cancer diseases, with limited private and public data sets and a lack of independent external model validation. 13 As a result, no AI applications have been implemented in clinical practice for PDAC.…”

“…4,19 These signs are often visible on CECT scans 18 to 12 months before clinical diagnosis, but the reported radiologists' sensitivity for their timely detection is only 44%, limiting the chances of early action. 19 Recent articles have shown great potential for AI-driven PDAC diagnosis in CT. 13 Chen et al 20 developed an algorithm to distinguish between pancreatic cancer and normal pancreas on portal venous CT, which was trained with a large data set of 2011 cases (546 pancreatic cancer). In a test set of 1473 CT studies (669 malignant) from institutions throughout Taiwan, AI achieved an area under the receiver operating characteristic curve (AUC) of 0.95 (95% confidence interval [CI] 0.94-0.96), with 74.7% (68 of 91, 95% CI 64.5-83.3) sensitivity for malignancies smaller than 2 cm.…”

“…21 Despite these promising early results, the identification of small lesions and secondary anatomic signs is still widely disregarded in AI-based detection research, and most studies do not disaggregate performance based on tumor size and stage. 13 In addition, there is a lack of research on lesion localization and a general absence of well curated data sets, with positive and negative cases being retrieved from completely different populations, which does not reflect the clinical landscape and can introduce bias. 13 For AI to improve PDAC detection, it is crucial to acquire and make publicly available well curated, multi-modal data sets that contain a significant proportion of small (<2 cm or even <1 cm) tumors, which should be treated as a subgroup of interest when reporting model performance.…”

“…Research in AI for histopathologic PDAC diagnosis is scarce. 13 Although histopathology is considered the criterion standard for confirming PDAC diagnosis, it is a timeconsuming process that suffers from non-uniform implementation in clinical practice and inter-observer variability. 13 Developing powerful AI models for histopathologic PDAC diagnosis is fundamental to advancing AI research at all steps of the patient pathway.…”