GenAI models score big in lung cancer grading and prognosis research

RadAI Slice Newsletter

Weekly Updates in Radiology AI

Good morning, there. Generative AI models achieved high accuracy in grading and prognosis of lung adenocarcinoma.

Generative AI tools are rapidly transforming digital pathology. By automating and standardizing the analysis of histological slides, they can outpace human experts in consistency, speed, and even unveil new prognostic insights. This progress highlights the expanding clinical impact of GenAI and hints at a coming wave of reproducible, resource-efficient cancer diagnostics.

Here’s what you need to know about Radiology AI last week:

• Generative AI transforms lung cancer grading and prognosis accuracy

• AI brings new consistency to IVF embryo health assessments

• LLMs shine in interpreting chest CT reports for thoracic disease

• New AI tool standardizes early detection across diverse diseases

• Plus: 6 FDA approved devices & 4 new papers.

🦠 Generative AI transforms lung cancer grading and prognosis accuracy

RadAI Slice: Generative AI models have set new benchmarks in digital pathology for lung adenocarcinoma grading and prognosis.

The details:

• Claude-3.5-Sonnet reached 82.3% accuracy for grading on a 492-slide dataset

• Models evaluated 11 histological and 4 clinical features for multi-factor risk assessment

• New prognostic factors, such as interstitial fibrosis, were identified by GenAI

• Tasks completed in minutes, offering objective, reproducible results

Key takeaway: GenAI can outperform pathologists in cancer grading, bringing major gains in standardization, efficiency, and outcome prediction in pathology workflows.

🧬 AI brings new consistency to IVF embryo health assessments

RadAI Slice: AI systems are boosting both the accuracy and agreement of embryologists evaluating IVF embryos, a pivotal IVF step.

The details:

• AI enables automated, objective embryo image analysis for IVF

• Meta-review of 37 studies finds AI outperforms manual assessment

• Deep learning analyzes all developmental stages using multimodal data

• Challenges remain in clinical workflow integration and ethics

Key takeaway: Imaging AI could raise the bar in fertility care by standardizing embryo evaluation and minimizing human variability—potentially improving IVF outcomes.

💡 LLMs shine in interpreting chest CT reports for thoracic disease

RadAI Slice: Large language models have proven effective as 'second readers' of chest CT radiology reports.

The details:

• Tested on 13,489 real-world chest CT reports covering 13 diseases

• GPT-4 achieved up to 75% accuracy with multiple-choice prompts

• Prompt tuning and fine-tuning further improved LLM performance

• No single LLM was best for all conditions, suggesting tailored deployment

Key takeaway: LLMs can significantly boost efficiency and accuracy in radiology report interpretation, reducing workload and error rates for radiologists.

🔬 New AI tool standardizes early detection across diverse diseases

RadAI Slice: Researchers at ECU unveil an AI model that enhances early disease detection with widely-used imaging scans.

The details:

• AI achieved 85% accuracy for early cardiovascular calcification detection

• Showed 87% accuracy for diabetic retinopathy, 91% for breast cancer staging

• Uses common modalities like ultrasound and bone density scans

• Will be presented at MICCAI 2025

Key takeaway: Accessible AI tools can offer accurate, standardized diagnoses across multiple conditions, enabling earlier and more equitable care in clinical practice.

🏛️ FDA Clearances

• K250901 - Canon's Vantage Fortian/Orian 1.5T MRI gets 510(k) with AiCE AI-based image reconstruction for superior MRI quality.

• K250686 - GyriCalc (NeuroSpectrum Insights) is cleared to quantify brain gyri patterns from MRI for neurological assessment.

• K251455 - Philips wins clearance for EPIQ/Affiniti ultrasound systems combining pulsed Doppler and live imaging for broad diagnostics.

• K250970 - Leo Cancer Care's Marie receives clearance as an advanced CT X-ray system for detailed cross-sectional imaging.

• K243681 - Olea Medical gains clearance for Neuro Insight V1.0, designed for advanced neurological image analysis.

• K251322 - GE's Venue series, now FDA-cleared, offers advanced pulsed Doppler ultrasound for varied clinical uses.

📄 Fresh Papers

• doi:10.1016/j.ejrad.2025.112339 - AI-based software improved malignancy risk prediction for breast lesions on targeted ultrasound in a single-center study.

• doi:10.1007/s11604-025-01839-w - Ultra-high-resolution CT approaches micro-CT standards, vastly improving trabecular bone imaging detail.

• doi:10.1016/j.ejmp.2025.105054 - Explainable radiomics classification of lung cancer subtypes was validated with biological and technical evidence.

• doi:10.1186/s12911-025-03120-6 - A multimodal deep learning model using carotid ultrasound and clinical data enables strong stroke risk prediction in hypertension.

📰 Everything else in Radiology AI last week

• NLP model identifies primary cancers from clinical notes, outperforming ICD codes and aiding targeted brain radiation.

• A new book synthesizes AI innovations in imaging, diagnosis, and care, emphasizing human empathy and explainability.

• Hyperfine's Swoop portable MRI, empowered by Optive AI, begins a trial to test effectiveness for emergency use.

• Experts predict AI in radiology will increase legal risks, impact patient ties, and prompt regulatory moves.

• EMBL's new AI-driven life sciences imaging plan gets major funding, expanding training, image processing, and ethics.

That's it for today!

👋 We’d love your thoughts. Just hit Reply and tell us what you want to see in the next Slice!

See you next week.