Deep learning

Papers on deep learning

2,582 papers208 new in 30 days

Thursday, November 19

Wednesday, November 18

Tuesday, November 17

bioRxiv

Model infers biological roles of unidentified molecules

In computational benchmarks, generated descriptions more closely matched reference descriptions than competing methods, without requiring the molecules behind the signals…In computational benchmarks, generated descriptions more closely matched reference descriptions than competing methods, without requiring the molecules behind the signals to be identified first.

Liver tissue with thick scars surrounding rounded cell clusters sits beside a smaller intact tissue fragment and scattered molecular fragments.
Preprint: computational studyAI & data

Monday, November 16

NEJM AI

Chest X-ray model ranked fracture risk better than standard tool

The model scored 0.843 versus 0.779 for a standard tool on a measure of how well it ranked patients’ fracture risk.The model scored 0.843 versus 0.779 for a standard tool on a measure of how well it ranked patients’ fracture risk.

Figure from NEJM AI
Cohort · 85,284 peopleMajor osteoporotic fractures
medRxiv

Automated tool measures muscle and fat from routine CT scans

In held-out scans, muscle area measurements averaged 1.35 square centimeters above radiologist-corrected labels at the third lumbar vertebra.In held-out scans, muscle area measurements averaged 1.35 square centimeters above radiologist-corrected labels at the third lumbar vertebra.

Abdominal cross-sections show thick versus thin dark blue-green muscles around gray tissue and a vertebra, beside small upright human figures.
Preprint: computational studyBiomarkers

Sunday, November 15

Saturday, November 14

European Heart Journal

Heart-recording model detects and predicts a heart muscle disease

Its scores for distinguishing people with and without the condition ranged from 0.87 to 0.97, outperforming traditional heart-recording measures.Its scores for distinguishing people with and without the condition ranged from 0.87 to 0.97, outperforming traditional heart-recording measures.

Diagnostic and predictive accuracy of AI-ECG model compared to a model based on traditional ECG and echocardiographic measurements for detection of prevalent and future DCM in general population and…
Cohort · 20,358 peopleBiomarkers

Friday, November 13

NeuroImage

Brain scans link older age to changes in fluid-producing tissue

In 234 people assessed for brain artery disease, advanced MRI detected more than twice the tissue volume found by a validated conventional MRI-based method.In 234 people assessed for brain artery disease, advanced MRI detected more than twice the tissue volume found by a validated conventional MRI-based method.

Figure from NeuroImage
Cross-sectional · 234 peopleMechanisms
European Radiology

AI shows promise for osteoporosis screening from X-rays

The model classified bone health with 80.1% accuracy in internal testing, while comparison models had lower observed accuracy.The model classified bone health with 80.1% accuracy in internal testing, while comparison models had lower observed accuracy.

A dark blue-green thighbone cutaway has sparse internal struts beside a smaller gray thighbone with dense struts and a vertebra with medium-sized gaps.
Cross-sectional · 854 peopleOsteoporosis