1. Automated analysis of the speech patterns of individuals showing early symptoms of psychosis predicted the later onset of the disorder with 93 percent accuracy, compared to clinicians’ 80 percent accuracy rate.
2. A group of 139 machine learning algorithms outperformed 511 healthcare professionals in diagnosing pigmented skin lesions from dermatoscopic images: The three highest-performing algorithms made an average of 25.43 correct diagnoses per 30-image set, while 27 human experts achieved fewer than 19 correct diagnoses per set.
3. Google’s AI team developed a deep learning algorithm that diagnosed lung cancer in CT scans with 11 percent fewer false positives and 5 percent fewer false negatives than radiologists.
4. A combination of machine learning and natural language processing used whole-genome sequencing to diagnose genetic diseases in children in a median time of about 20 hours, significantly faster than time-consuming manual interpretation.
5. A set of AI systems crowdsourced from data scientists segmented lung tumors with equal accuracy to radiation oncologists, but significantly faster: The algorithms completed the task between 15 seconds and two minutes, compared to the average eight minutes spent on it by humans.
6. Researchers created a deep learning algorithm that detected cervical precancer and cancer from archived photos of cervixes with greater accuracy than the human clinicians that had visually inspected the cervixes.
7. Software company Eko’s AI analyzed heart sounds data gathered by noninvasive sensors to outperform four out of five cardiologists in detecting pediatric heart murmurs.
More articles about AI:
5 key quotes about how AI will transform healthcare
Harvard Business Review: 3 shifts organizations must make to scale AI
Cleveland Clinic uses AI, medical records, scans to personalize radiation therapy
