AI may predict PGL gene cluster
An AI–based analysis of reticulin architecture may help predict molecular cluster status in paragangliomas (PGLs), according to a study published in Endocrine Pathology. The study evaluated whether histoarchitectural features visible on routine reticulin stain correlate with underlying germline genotype.
Pheochromocytomas and extra-adrenal paragangliomas are neuroendocrine neoplasms with a high rate of heritable mutations. Molecular classification stratifies these tumors into clusters with distinct biological and clinical features. Cluster 1 tumors, associated with pseudohypoxic pathways and genes such as SDHx and VHL, typically present at a younger age and may demonstrate different biochemical profiles compared with cluster 2 tumors, according to Eleonora Duregon, MD, PhD, of the University of Turin, Italy, and colleagues.
The investigators retrospectively analyzed 104 surgically resected PGLs with complete clinical and germline genetic data. The cohort included 90 pheochromocytomas and 14 extra-adrenal paragangliomas. Germline mutations were identified in 39 patients (37.5%): 20 with cluster 1 variants and 19 with cluster 2 variants. The remaining 65 tumors were classified as sporadic.
Reticulin staining revealed two main architectural patterns: intact framework and disrupted framework. A subset of intact cases showed a "very small nest pattern," defined as nests composed of one to four cells. Intact reticulin was present in 51.9% of tumors overall but was significantly more frequent in cluster 1 tumors (85%) than in cluster 2 or sporadic tumors. Very small nests were also significantly enriched in cluster 1 tumors (p=0.003).
To standardize evaluation, the authors developed a convolutional neural network trained on digitized whole-slide images of reticulin-stained sections. The model segmented tumor tissue, quantified intact reticulin areas, and identified very small nests. Validation demonstrated high precision and sensitivity for both features.
Quantitatively, the median tumor area composed of very small nests was 35.2% in cluster 1 tumors, compared with 7.3% in cluster 2 and 9.4% in sporadic tumors. Median intact reticulin area was 53% in cluster 1 tumors versus approximately 30% in the other groups.
Two bias-reduced logistic regression models were developed to predict cluster 1 genotype using age, tumor size, extra-adrenal location, and either percentage intact reticulin area or percentage of very small nests. The intact reticulin model achieved an area under the receiver operating characteristic curve (AUC) of 0.981. The very small nest model achieved an AUC of 0.990 after bootstrap validation.
Younger age, larger tumor size, extra-adrenal presentation, and higher AI-derived reticulin metrics were associated with increased probability of cluster 1 genotype.
"From a clinical standpoint, our data suggest two practical implications. First, routine reticulin staining can provide additional morphological context when evaluating PGLs with suspected pseudohypoxic background, potentially prioritizing cases for genetic counseling/testing. Second, AI-assisted morphometrics offers a scalable, observer-independent tool to standardize this readout and integrate it into digital workflows, which may be particularly valuable in resource-constrained settings," wrote study investigators.
"These insights advocate for bridging conventional histopathology with computational analysis to surface latent diagnostic signals in routine stains, refine pre-test probability for genetics, and ultimately support personalized care pathways in PGLs," concluded Dr. Duregon and colleagues.
The study was limited by its single-center design and absence of somatic mutation analysis. External validation will be required before clinical adoption.
Several investigators reported tied to pharmaceutical companies, including AbbVie, AstraZeneca, and Eli Lilly.
AACE Endocrine AI is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (AACE®). The ideas and opinions expressed in AACE Endocrinology do not necessarily reflect those of Conexiant or AACE. For more information, see Policies.
Performance of a convolutional neural network in determining differentiation levels of cutaneous squamous cell carcinomas was on par with that of experienced dermatologists, according to the results of a recent study published in JAAD International.
“This type of cancer, which is a result of mutations of the most common cell type in the top layer of the skin, is strongly linked to accumulated [ultraviolet] radiation over time. It develops in sun-exposed areas, often on skin already showing signs of sun damage, with rough scaly patches, uneven pigmentation, and decreased elasticity,” stated lead researcher Sam Polesie, MD, PhD, Associate Professor of Dermatology and Venereology at the University of Gothenburg and Practicing Dermatologist at Sahlgrenska University Hospital, both in Gothenburg, Sweden.