Latest Research
All publications from the Cancer3.AI database, newest first.
Machine learning and deep learning models for predicting colorectal cancer metastases: A comprehensive review.
Getu MA, et al
This comprehensive review examined the current landscape of machine learning (ML) and deep learning (DL) models developed to predict the spread of colorectal cancer (CRC) to distant sites, including lymph nodes, liver, lungs, bones, and the peritoneum. Researchers systematically evaluated a range of approaches, from traditional algorithms such as logistic regression and random forests to advanced deep learning architectures including convolutional neural networks like GoogleNet, VGGNet, ResNet, and U-Net, which can identify complex patterns across imaging, clinical, histological, and molecular data. The review found that DL models incorporating radiomics and transfer learning consistently achieved superior predictive performance compared to conventional methods, and that integrating multi-modal data further enhanced accuracy in identifying patients at risk of metastasis. These advances hold significant promise for personalizing cancer treatment strategies, enabling earlier intervention and potentially improving survival rates for colorectal cancer patients worldwide. However, the authors identified critical challenges that must be addressed before widespread clinical adoption, including the need for high-quality diverse datasets, improved model interpretability, sustainable computational resources, and robust ethical frameworks. Future research priorities include developing explainable AI models and validating predictions across diverse patient populations to ensure equitable and reliable clinical application.
European journal of radiology open
Source →Clear Cell Renal Cell Carcinoma: Rare and Atypical Metastatic Localizations-A Case Series.
Tessa Djambong R, et al
A new case series published in Case Reports in Urology documents four patients with clear cell renal cell carcinoma (ccRCC) — the most common subtype of kidney cancer — who developed metastases in rare and unexpected anatomical locations. The study reports metastatic spread to the nasal cavity, thyroid gland, gallbladder, and pancreas, with pancreatic involvement observed in all four patients. These unusual metastatic sites were frequently asymptomatic, contributing to diagnostic delays and highlighting the challenge of detecting recurrence beyond typical organs such as the lungs, bones, and liver. Each patient received individualized treatment — including active surveillance, surgical resection, radiotherapy, or immunotherapy — reflecting the necessity of tailored management strategies in this complex disease. The authors stress that patients who have undergone nephrectomy for ccRCC require prolonged and personalized follow-up, as metastases can emerge years later in atypical locations. This case series underscores the essential role of multidisciplinary teams in ensuring timely detection and optimal care for patients with rare metastatic presentations of kidney cancer.
Case reports in urology
Source →Impact of Minimum Segment Width on VMAT Plan Quality for Nasopharyngeal Carcinoma.
Li Z, et al
Researchers investigated how a key technical parameter in radiotherapy planning software — the minimum segment width (MSW) — affects treatment plan quality for patients with nasopharyngeal carcinoma (NPC), a cancer arising in the upper throat region behind the nose. Using planning CT scans from 40 NPC patients across all tumor stages, four separate volumetric modulated arc therapy (VMAT) plans were generated for each patient with MSW values of 0.5, 0.8, 1.0, and 1.5 cm, while all other planning parameters remained identical. For early-stage (T1-2) patients, smaller MSW values produced superior dose conformity and more homogeneous coverage of target volumes, while also delivering lower radiation doses to critical nearby structures such as the parotid glands. For advanced-stage (T3-4) patients, none of the plans achieved the desired 95% target coverage threshold, but smaller MSW values still offered meaningful improvements in target dose and sparing of healthy tissues including the parotid glands, jaw joints, and oral mucosa. Although smaller MSW values increased monitor units and thus treatment delivery time, the overall evidence points to MSW settings of 0.5 to 0.8 cm as optimal for NPC radiotherapy planning, offering clinicians a practical, evidence-based reference for improving patient outcomes.
Dose-response : a publication of International Hormesis Society
Source →Transoral Endoscopic Plasma Resection for Diagnosis and Management of Laryngeal Inflammatory Myofibroblastic Tumor: A Case Report.
Luo W, et al
Inflammatory myofibroblastic tumor (IMT) is a rare, typically non-malignant neoplasm that almost never arises in the larynx, making its clinical recognition and surgical management particularly challenging. This case report describes a 24-year-old woman who presented with breathing difficulties and was found, via cervical computed tomography, to have a laryngeal mass consistent with IMT. Surgeons successfully performed transoral endoscopic plasma resection under general anesthesia, a minimally invasive technique that removes laryngeal growths entirely through the mouth, avoiding external incisions. Postoperative histopathological analysis confirmed the diagnosis of IMT, and the patient recovered fully with no evidence of tumor recurrence during follow-up. The case demonstrates that transoral endoscopic plasma resection is a safe, feasible, and effective approach for both diagnosing and treating this uncommon laryngeal neoplasm. These findings offer practical guidance for otolaryngologists who may encounter rare laryngeal tumors requiring precise, function-preserving surgical intervention.
Ear, nose, & throat journal
Source →CKAD-Net: class-guided key-feature aggregation and adaptive decision network for vocal cord lesion prediction.
Liu Y, et al
Researchers developed CKAD-Net, a deep learning neural network designed to automatically detect and classify vocal cord lesions from laryngoscopy images. The model introduces two key innovations: a category-guided key feature aggregation mechanism that adaptively focuses attention on the most diagnostically relevant visual features for each lesion type, and an adaptive decision mechanism that uses learnable weighting to dynamically combine predictions from multiple feature representations. When evaluated on the hospital-private dataset VCLScopeData, CKAD-Net achieved an AUC of 0.944 and accuracy of 86.4% for coarse-grained lesion classification, and an AUC of 0.929 with 73.0% accuracy for the more challenging fine-grained task, outperforming existing state-of-the-art models. These results are clinically significant because undiagnosed vocal cord lesions can progress to laryngeal cancer, and an AI tool capable of reliable early classification could help clinicians prioritize high-risk patients and reduce diagnostic delays. This work represents a meaningful advance toward AI-assisted laryngoscopy that may ultimately improve voice preservation and patient survival outcomes.
Scientific reports
Source →