Latest Research
All publications from the Cancer3.AI database, newest first.
Kaposi sarcoma herpesvirus (KSHV) subtypes and impact on outcomes in KSHV-associated diseases.
Matos JM, et al
Researchers conducted a retrospective study examining whether genetic subtypes of Kaposi sarcoma-associated herpesvirus (KSHV) — a virus that causes several serious diseases in people living with HIV — influence disease presentation and survival outcomes. The study analyzed 143 HIV-positive patients treated between 2004 and 2024, sequencing the viral K1 gene to classify KSHV into genetic subtypes and linking these subtypes to clinical diagnoses including Kaposi sarcoma, multicentric Castleman disease, primary effusion lymphoma, and KSHV inflammatory cytokine syndrome. Overall, KSHV subtype A was the most common, found in 46% of patients, but subtypes did not differ by disease type and had no significant impact on survival across the broader patient group. A clinically important exception emerged: patients with primary effusion lymphoma who carried subtype A had dramatically worse survival compared to those with other subtypes, with a median overall survival of 1.6 years versus 7.2 years. These findings suggest that while KSHV genetic variation generally does not drive disease outcomes, subtype A may play a specific and harmful role in the most aggressive form of KSHV-associated lymphoma. The authors call for further research to understand the biological mechanisms by which KSHV genetic diversity may affect disease development and treatment response.
The Journal of infectious diseases
Source →Multiview 2.5D Deep Learning Outperforms 2D and 3D Models for Preoperative Prediction of Visceral Pleural Invasion in Stage IA Lung Adenocarcinoma.
Zhao J, et al
Researchers from two Chinese hospitals studied whether a novel artificial intelligence approach called 2.5-dimensional (2.5D) deep learning could better predict visceral pleural invasion (VPI) — a key factor affecting surgical planning — in patients with early-stage lung adenocarcinoma. The study analyzed CT scans from 804 patients and compared four types of deep learning models: standard 2D (single-slice), 3D (volumetric), and two new 2.5D approaches that combine information from multiple imaging planes. The 2.5D multiview model, which integrated the largest tumor cross-sections from three anatomical planes simultaneously, achieved the highest accuracy with an AUC of 0.73 in external testing, outperforming all other approaches. Importantly, the model also provided visual explanations via Grad-CAM heat maps, highlighting the tumor-pleura contact zone as the most critical region for predicting invasion. These findings matter clinically because accurately identifying VPI before surgery helps surgeons choose the appropriate extent of lung resection, directly affecting patient outcomes and staging. This is the first study to validate a 2.5D multiview deep learning model for this specific task, offering a promising tool for non-invasive, AI-assisted preoperative decision-making in lung cancer care.
Journal of thoracic imaging
Source →INTELCAPE: A Deep Learning-Powered System for Automated, High-Accuracy Crohn's Disease Diagnosis via Capsule Endoscopy.
Fan D, et al
Researchers developed INTELCAPE, an artificial intelligence system designed to automate the analysis of capsule endoscopy videos for diagnosing Crohn's disease, a chronic inflammatory bowel condition. The system was trained and tested on data from 757 and 115 patients across two Chinese hospitals, using a multi-task deep learning framework combining ResNet, Transformer, and EfficientNet architectures to segment intestinal regions, detect lesions, and deliver diagnoses. INTELCAPE achieved lesion classification accuracy of 99.33% and diagnostic accuracy of 90%, performance comparable to specialist physicians, while operating ten times faster than human reviewers. Crucially, when doctors used INTELCAPE as an assistive tool, their diagnostic accuracy improved from 76.7% to 94.8%, and their video interpretation time dropped from nearly 68 minutes to just 22.5 minutes. These findings suggest that AI-assisted capsule endoscopy analysis could substantially reduce the burden on clinicians and improve diagnostic quality, especially for less experienced practitioners. INTELCAPE represents a significant step toward integrating deep learning into routine gastrointestinal diagnostics.
Clinical gastroenterology and hepatology : the official clinical practice journal of the American Gastroenterological Association
Source →Long-Term Clinical Consequences of Severe Oral Mucositis in Survivors of Lip, Oral Cavity, and Pharynx Cancer Versus Leukemia: A Propensity-Score-Matched Comparative Cohort Study Using Real-World Data.
Satheeshkumar PS, et al
A large population-based study using real-world data from over 110 million patients examined whether severe oral mucositis — painful mouth sores caused by cancer treatment — leaves lasting health consequences that differ depending on the type of cancer treated. Researchers compared long-term outcomes in survivors of lip, oral cavity, and pharynx cancers (CLOP) versus leukemia, using propensity-score matching to ensure fair comparisons between patients who did and did not develop severe mucositis. In head and neck cancer survivors, severe mucositis was associated with nearly double the lifetime mortality risk, more than triple the risk of swallowing difficulties, and substantially elevated risks of malnutrition, respiratory disease, and pneumonia. By contrast, leukemia survivors who experienced mucositis showed only modest increases in the same outcomes, suggesting the structural damage to the upper digestive and airway tract in head and neck cancer patients drives these lasting harms. These findings challenge the widespread assumption that oral mucositis is merely a temporary side effect, and indicate that head and neck cancer survivors who develop severe mucositis should receive long-term surveillance and proactive management of swallowing function, nutrition, and lung health.
Medical sciences (Basel, Switzerland)
Source →Neuroprotective Role of DING Protein in Normal Aging and Alzheimer's Disease.
Darbinian N, et al
Researchers investigated the neuroprotective potential of the DING protein — a phosphatase originally isolated from St. John's wort — in human brain tissue affected by Alzheimer's disease and in normal aging. The study compared postmortem brain samples from five patients with dementia, including three with confirmed Alzheimer's neuropathology, against five non-demented controls, while also using PC12 rat neuronal cells to examine how DING influences the phosphorylation of the Tau protein. Key findings revealed that DING protein was present in neuronal cell bodies and processes in both healthy and Alzheimer's-affected brain tissue, and that it demonstrated phosphatase activity capable of inhibiting Tau phosphorylation — a process that, when excessive, causes the formation of neurofibrillary tangles, a hallmark of Alzheimer's neurodegeneration. In brain samples from individuals aged 72 to 92 years, higher levels of endogenous DING protein correlated positively with reduced Tau phosphorylation, and increasing DING expression in neuronal cells improved cell survival. These findings suggest that DING protein could represent a promising therapeutic target for slowing or preventing neurodegeneration in Alzheimer's disease.
Archives of internal medicine research
Source →