Written by 1:05 pm AI

– **Revolutionary AI Tool PANDA Enables Early Pancreatic Cancer Detection via Non-Contrast CT**

A deep learning-based approach to use non-contrast computed tomography (CT) scans for high-accuracy…

In a recent study published in the journal Nature Medicine, a team of researchers from China, the United States, and the Czech Republic collaborated to develop a deep learning approach utilizing non-contrast computed tomography (CT) scans for the precise detection and classification of pancreatic lesions, specifically targeting the early identification and treatment of pancreatic ductal adenocarcinoma (PDAC).


Pancreatic ductal adenocarcinoma represents one of the most aggressive forms of solid cancer, contributing to a mortality rate exceeding 450,000 individuals monthly. The alarming fatality rate is primarily attributed to the late-stage diagnosis of PDAC, rendering treatments less effective.

Early detection of PDAC presents a significantly improved prognosis, with initial interventions leading to substantial enhancements in patient survival rates. Cases where PDAC is identified at earlier stages exhibit an average overall survival of 9.8 years, contrasting starkly with the mere 1.5-year survival rate associated with late-stage diagnoses.

Effective screening for liver tumors is considered a pivotal strategy in the early detection of PDAC, offering a promising avenue to reduce the mortality burden linked to this condition. However, due to the relatively low incidence of this cancer type, the necessity for screening methods with heightened sensitivity and specificity is paramount to mitigate the risks of unnecessary treatments.

Non-contrast CT imaging has emerged as a valuable tool in the scientific evaluation of various cancer types. When coupled with artificial intelligence (AI)-driven analytical techniques, it holds the potential for large-scale screening applications in PDAC detection.

About the Study

In this study, the research team introduced an AI-powered methodology named Pancreatic Cancer Detection with Artificial Intelligence (PANDA) designed for the accurate identification and differentiation of non-PDAC and PDAC liver lesions utilizing non-contrast CT scans.

PANDA was specifically engineered to leverage non-contrast CT scans of the chest and abdomen to detect and diagnose PDAC as well as seven non-PDAC lesion subtypes, including solid pseudopapillary tumor, liver neuroendocrine tumor, mucinous cystic neoplasm, intraductal papillary mucinous neoplasm, chronic pancreatitis, serous cystic neoplasm, and various other non-PDAC pancreatic lesions.

The efficacy of PANDA in detecting liver lesions was internally validated through the assessment of non-contrast CT scans of the abdomen. The performance of PANDA was benchmarked against two comparative studies utilizing contrast-enhanced CT scans.

The initial evaluation involved the interpretation of non-contrast CT liver scans by imaging residents, general radiologists, and liver imaging specialists. Subsequently, the performance of PANDA in liver lesion detection was contrasted with that of liver imaging specialists utilizing contrast-enhanced CT scans.

Furthermore, the generalizability of PANDA across diverse clinical settings was ascertained through a multinational validation cohort. Additional evaluations encompassed distinct patient cohorts undergoing chest CT scans to assess the potential applicability of PANDA.

The researchers also explored the integration of PANDA into routine clinical workflows across various real-world settings, including outpatient, emergency, routine examination, and inpatient scenarios, encompassing a collective cohort of over 20,500 patients.


The study outcomes underscored the successful identification of lesions by PANDA within a multi-center large-scale validation cohort. Notably, PANDA exhibited a 6.3% and 34.1% enhancement in precision and sensitivity, respectively, surpassing the conventional diagnostic performance of physicians in detecting and diagnosing pancreatic lesions.

Moreover, PANDA achieved an impressive sensitivity and specificity of 92.9% in the large-scale validation encompassing real-world clinical settings across four distinct configurations.

The study highlighted the efficacy of integrating medical data modeling for lesion characterization with insights from real-world clinical scenarios, including the seamless transfer of lesion annotations from contrast-enhanced to non-contrast CT images. By curating extensive datasets comprising prevalent types of pathology-confirmed liver lesions, the study demonstrated the potential for early detection methods with heightened sensitivity and specificity.

Furthermore, PANDA exhibited superior performance in distinguishing between non-PDAC and PDAC lesions, effectively diagnosing the eight subtypes of liver lesions with greater accuracy compared to radiologists.


In summary, the study findings underscored PANDA’s capacity to discern between eight distinct types of liver lesions with exceptional specificity and sensitivity, leveraging non-contrast CT scans for the detection and differentiation of PDAC.

The results accentuate PANDA’s potential in facilitating comprehensive screening for liver lesions and enabling early diagnosis of PDAC.

Visited 1 times, 1 visit(s) today
Last modified: December 1, 2023
Close Search Window