Researchers tested a large language model (LLM) on peer review tasks for cancer research papers. They found the AI could be abused to generate highly persuasive rejection letters and other fraudulent reviews, such as requests to cite unrelated papers. Crucially, current AI detection tools were largely unable to identify the AI-generated text, posing a significant, hidden threat to academic integrity.

The objective of this study is to assess the diagnostic performance of image analysis-capable generative AI (Gen-AI) (GPT-4-turbo, Google DeepMind's Gemini-pro-vision, and Anthropic’s Claude-3-opus) in interpreting CT images of lung cancer. This is the first study to integrate the diagnostic capabilities of these three models across distinct imaging settings. Additionally, a Likert scale is used to evaluate each model's internal tendencies. By examining the potential and limitations of multimodal large language models (MM-LLMs) for lung cancer diagnosis, this research aims to provide an evidence-based foundation for the future clinical applications of Gen-AI.

Researchers have successfully demonstrated that advanced generative AI (GenAI) models can accurately assess lung adenocarcinoma pathological features with remarkable precision. The comprehensive study shows Claude-3.5-Sonnet achieving 82.3% accuracy in cancer grading, potentially revolutionizing how pathologists diagnose and predict outcomes for lung cancer patients.

Researchers have developed a compact, noninvasive imaging system that combines high-resolution structural imaging with chemical analysis to improve skin cancer diagnosis. The system integrates line-field confocal optical coherence tomography and confocal Raman microspectroscopy, allowing clinicians to examine both the cellular structure and molecular composition of skin tissues. In a year-long clinical study involving over 330 nonmelanoma skin cancer samples, the system enabled targeted chemical analysis of suspicious structures. An AI model trained on the spectral data achieved high accuracy in identifying cancerous tissues, with classification scores of 0.95 for basal cell carcinoma and 0.92 when including squamous cell carcinoma. This dual-modality approach promises to enhance diagnostic precision and deepen understanding of skin cancer biology.

A research team from Southern Medical University has developed a machine learning-based gene model that predicts whether nasopharyngeal cancer (NPC) patients will benefit from radiotherapy. This predictive tool, called the NPC-RSS, was validated in both cell lines and patient samples. The model may guide personalized treatment decisions and improve survival outcomes for NPC patients.