Clinical problem

Pancreatic cancer is estimated to become the second leading cause of cancer-related deaths in Western countries by 2030. Pancreatic ductal adenocarcinoma (PDAC), the most common and aggressive type of pancreatic cancer, cannot be effectively prevented or screened for and is associated with 98% life expectancy loss. Due to the lack of early, disease-specific symptoms, 80–85% of patients are diagnosed in advanced disease stages. However, patients at stage I present a significantly more favorable prognosis than stage IV patients (median survival: 26 vs. 4.8 months), making early detection the current most effective approach to improving outcome (Schwartz et al., 2021).

Abdominal contrast-enhanced computed tomography (CECT) scans are usually the first line of diagnosis for PDAC, as there are no validated early diagnostic biomarkers. Studies have shown that in 16%-84% of cases cancer signs (such as pancreatic duct cutoff/dilatation and pancreatic atrophy) can be retrospectively seen on CECT scans 3–36 months before clinical diagnosis (Singh et al., 2020; Toshima et al., 2021). These secondary signs and subtle focal lesions can be overlooked in clinical practice, where most abdominal CECT scans are acquired for non-pancreatic-specific indications, leading to delayed diagnosis and reduced survival.

Artificial intelligence (AI) is starting to achieve expert performances in cancer diagnosis across many domains (Esteva et al., 2017; McKinney et al., 2020). The gold standard for assessing and benchmarking AI algorithms is international competitions or \xe2\x80\x9cgrand challenges\xe2\x80\x9d, where algorithm developers compete against each other using common datasets and well-defined performance metrics. An increasing number of publications study AI for PDAC diagnosis on CECT (Korfiatis et al., 2023; Chen et al., 2023; Park et al., 2023). However, there are currently no AI grand challenges for PDAC diagnosis, and every study evaluates its AI algorithms using private data sets, usually originating from a single center. This limits AI generalizability and makes it impossible to compare different approaches reliably. Furthermore, most studies do not focus on early detection or stratify AI results based on tumor size and stage.

The PANORAMA challenge

The PANORAMA (Pancreatic cancer diagnosis: Radiologists meet AI) study is a new prospectively designed multi-center study with over 1500 cases, established in conjunction with an international, multi-disciplinary scientific advisory board (11 experts in pancreas radiology, AI and pancreatic cancer survivor representative) \xe2\x81\xa0\xe2\x80\x94to unify and standardize present-day guidelines and to ensure meaningful validation of pancreas-AI towards clinical translation (Reinke et al., 2022).

Within the PANORAMA study, we will host the first international reader study and AI grand challenge for PDAC detection on CECT, where we will assess radiologists and AI algorithms from all over the world using a standardized and controlled environment.

The PANORAMA study primarily consists of two sub-studies:

• AI study: An annotated data set consisting of retrospective, multi-center, cancer-enriched routine upper-abdominal CECT scans. The data set aims to represent clinical reality by including diagnostic and pre-diagnostic PDAC scans as well as non-PDAC patients (both healthy pancreas patients and patients with non-PDAC pancreatic lesions). Teams can use this dataset to develop AI models, and submit their trained algorithms (in Docker containers) for evaluation. At the end of this open development phase, all algorithms are ranked, based on their performance on a hidden testing cohort of 400 unseen scans.

• Reader study: The PANORAMA reader study will include around\xe2\x80\xaf400 cases\xe2\x80\xaffrom multiple institutions that will be read by a panel of\xe2\x80\xaf40+ international readers\xe2\x80\xafwith varying levels of expertise.

In the end, PANORAMA aims to benchmark state-of-the-art AI algorithms developed in the grand challenge, against abdominal radiologists participating in the reader study \xe2\x80\x94to evaluate the clinical viability of modern pancreas-AI solutions at PDAC detection and diagnosis in CECT.

Partners & Organizers

‘]