We trained and evaluated a localization-based deep CNN for breast cancer screening exam classification on over 200,000 exams (over 1,000,000 images). Our model achieves an AUC of 0.919 in predicting malignancy in patients undergoing breast cancer screening, reducing the error rate of the baseline (Wu et al., 2019a) by 23%. In addition, the models generates bounding boxes for benign and malignant findings, providing interpretable predictions.
Recent American news events range from horrifying to dystopian, but reading the applications of our fast.ai I was blown away by how many bright, creative, resourceful folks from all over the world are applying deep learning to tackle a variety of meaningful and interesting problems. Their passions range from ending illegal logging, diagnosing malaria in rural Uganda, translating Japanese manga, reducing farmer suicides in India via better loans, making Nigerian fashion recommendations, monitoring patients with Parkinson's disease, and more. Our mission at fast.ai is to make deep learning accessible to people from varied backgrounds outside of elite institutions, who are tackling problems in meaningful but low-resource areas, far from mainstream deep learning research. Our group of selected fellows for Deep Learning Part 2 includes people from Nigeria, Ivory Coast, South Africa, Pakistan, Bangladesh, India, Singapore, Israel, Canada, Spain, Germany, France, Poland, Russia, and Turkey.
Artificial intelligence is enabling many scientific breakthroughs, especially in fields of study that generate high volumes of complex data such as neuroscience. As impossible as it may seem, neuroscientists are making strides in decoding neural activity into speech using artificial neural networks. Yesterday, the neuroscience team of Gopala K. Anumanchipalli, Josh Chartier, and Edward F. Chang of University of California San Francisco (UCSF) published in Nature their study using artificial intelligence and a state-of-the-art brain-machine interface to produce synthetic speech from brain recordings. The concept is relatively straightforward--record the brain activity and audio of participants while they are reading aloud in order to create a system that decodes brain signals for vocal tract movements, then synthesize speech from the decoded movements. The execution of the concept required sophisticated finessing of cutting-edge AI techniques and tools.
WASHINGTON D.C. [USA]: According to a recent study, a new artificial intelligence technology can accurately identify rare genetic disorders using a photograph of a patient's face. Named DeepGestalt, the AI technology outperformed clinicians in identifying a range of syndromes in three trials and could add value in personalised care, CNN reported. The study was published in the journal Nature Medicine. According to the study, eight per cent of the population has disease with key genetic components and many may have recognisable facial features. The study further adds that the technology could identify, for example, Angelman syndrome, a disorder affecting the nervous system with characteristic features such as a wide mouth with widely spaced teeth etc. Speaking about it, Yaron Gurovich, the chief technology officer at FDNA and lead researcher of the study said, "It demonstrates how one can successfully apply state of the art algorithms, such as deep learning, to a challenging field where the available data is small, unbalanced in terms of available patients per condition, and where the need to support a large amount of conditions is great."
The Centers for Disease Control and Prevention (CDC) coordinates a labor-intensive process to measure the prevalence of autism spectrum disorder (ASD) among children in the United States. Random forests methods have shown promise in speeding up this process, but they lag behind human classification accuracy by about 5 percent. We explore whether newer document classification algorithms can close this gap. We applied 6 supervised learning algorithms to predict whether children meet the case definition for ASD based solely on the words in their evaluations. We compared the algorithms? performance across 10 random train-test splits of the data, and then, we combined our top 3 classifiers to estimate the Bayes error rate in the data. Across the 10 train-test cycles, the random forest, neural network, and support vector machine with Naive Bayes features (NB-SVM) each achieved slightly more than 86.5 percent mean accuracy. The Bayes error rate is estimated at approximately 12 percent meaning that the model error for even the simplest of our algorithms, the random forest, is below 2 percent. NB-SVM produced significantly more false positives than false negatives. The random forest performed as well as newer models like the NB-SVM and the neural network. NB-SVM may not be a good candidate for use in a fully-automated surveillance workflow due to increased false positives. More sophisticated algorithms, like hierarchical convolutional neural networks, would not perform substantially better due to characteristics of the data. Deep learning models performed similarly to traditional machine learning methods at predicting the clinician-assigned case status for CDC's autism surveillance system. While deep learning methods had limited benefit in this task, they may have applications in other surveillance systems.