Industry Insights AI
This year’s second round of insights highlights three of the latest advancements of the application of artificial intelligence (AI) in healthcare.
AI ultrasound models can outperform sonographers in ovarian cancer diagnosis
A recent meta-analysis published in Frontiers in Oncology found that AI models analysing ultrasound images achieved higher sensitivity and specificity for ovarian cancer detection than experienced sonographers. Internal validation showed an area under the curve (AUC) of 0.98; external validation also maintained strong performance. These results showcase the potential of AI models in improving early diagnosis, reducing unnecessary surgeries, and supporting clinical decision-making. However, the authors emphasised the need for prospective, multicentre studies before routine clinical adoption of these models.
Full article: https://shorturl.at/UeNkN
AI wearables can reduce nocturnal scratching in atopic dermatitis
A recent study published in JAMA Dermatology reported that an AI-powered wearable sensor significantly reduced nighttime scratching in adults with mild atopic dermatitis. Atopic dermatitis is a common skin condition that causes itchy, red, dry and inflamed skin. The device delivered real-time haptic feedback and demonstrated a 99% accuracy in detecting scratch events. Over two weeks, participants experienced a 28% reduction in scratch events and a 40% reduction in total scratch duration per night. These findings suggest AI-enabled wearables could provide a non-pharmacologic option to interrupt the itch–scratch cycle, particularly for patients who do not require systemic therapy.
Article abstract: https://shorturl.at/HVy8g
DeepSeek-R1 outperforms ChatGPT-4o in urticaria-related questions
A comparative study evaluating large language models found that DeepSeek-R1 consistently outperformed ChatGPT-4o when responding to urticaria-related questions. Dermatologists and non-specialists rated DeepSeek higher than its counterpart for accuracy, guideline adherence and clarity. DeepSeek demonstrated no guideline-contradicting errors. However, both generative AI (genAI) models showed limitations in deep knowledge. The authors stressed that the model requires regular updating and validation before clinical integration.
Full article: https://shorturl.at/NUiXz
