Our new preprint implementing a deep learning-based approach for automatic carotid plaque detection in ultrasound imaging!

📢 𝐅𝐫𝐞𝐬𝐡 𝐨𝐮𝐭 𝐟𝐫𝐨𝐦 𝐨𝐮𝐫 𝐥𝐚𝐛: A new preprint on 𝐠𝐞𝐧𝐞𝐭𝐢𝐜𝐬 𝐨𝐟 𝐜𝐚𝐫𝐨𝐭𝐢𝐝 𝐚𝐭𝐡𝐞𝐫𝐨𝐬𝐜𝐥𝐞𝐫𝐨𝐬𝐢𝐬 and 𝐝𝐞𝐞𝐩 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠 for automating plaque detection in vascular ultrasound❗

We analyzed 180,000 images from the population-based  𝐔𝐊 𝐁𝐢𝐨𝐛𝐚𝐧𝐤 and developed a deep learning model for automatic plaque detection in carotid ultrasound 📊

Carotid plaques are a well-established marker of 𝐬𝐮𝐛𝐜𝐥𝐢𝐧𝐢𝐜𝐚𝐥 𝐚𝐭𝐡𝐞𝐫𝐨𝐬𝐜𝐥𝐞𝐫𝐨𝐬𝐢𝐬 and may be present long before acute cardiovascular events, such a stroke or a myocardial infarction. Ultrasound is a widely accessible, safe, and accurate method for detecting plaques.

👉our model achieved 𝐡𝐢𝐠𝐡 𝐩𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐦𝐞𝐭𝐫𝐢c𝐬 with accuracy, sensitivity and specificity of 90%
👉 applying the model in 20,000 participants,  𝐰𝐞 𝐟𝐨𝐮𝐧𝐝 𝐩𝐥𝐚𝐪𝐮𝐞𝐬 𝐢𝐧 𝟒𝟓% 𝐨𝐟 𝐭𝐡𝐞 𝐩𝐨𝐩𝐮𝐥𝐚𝐭𝐢𝐨𝐧 (aged 47-83 years)
👉 plaque presence and count were 𝐫𝐨𝐛𝐮𝐬𝐭 𝐩𝐫𝐞𝐝𝐢𝐜𝐭𝐨𝐫𝐬 of future myocardial infarction, stroke, or vascular death
👉 adding plaque presence or count to established clinical prediction models 𝐫𝐞𝐜𝐥𝐚𝐬𝐬𝐢𝐟𝐢𝐞𝐝 𝟔% 𝐨𝐟 𝐩𝐚𝐫𝐭𝐢𝐜𝐢𝐩𝐚𝐧𝐭𝐬 who went on to experience cardiovascular events to a higher risk category, which would make them eligible for preventive statin treatment
👉 leveraging the unique omics data in UK Biobank, we ran the 𝐥𝐚𝐫𝐠𝐞𝐬𝐭 𝐭𝐨-𝐝𝐚𝐭𝐞 𝐆𝐖𝐀𝐒 for carotid atherosclerosis (29,790 cases & 36,847 controls)
👉 we detected 𝐭𝐰𝐨 𝐧𝐨𝐯𝐞𝐥 𝐠𝐞𝐧𝐨𝐦𝐢𝐜 𝐥𝐨𝐜𝐢 influencing risk of carotid plaques
👉downstream analyses including drug target Mendelian randomization revealed significant signals for  𝐋𝐩(𝐚) 𝐚𝐧𝐝 𝐈𝐋-𝟔 𝐬𝐢𝐠𝐧𝐚𝐥𝐢𝐧𝐠, both targets of investigational drugs in advanced clinical development for cardiovascular disease

Great work by star PhD student Murad Omarov ❗

🔗 https://www.medrxiv.org/content/10.1101/2024.10.17.24315675v1