Non-small cell lung cancer diagnosis aid with histopathological images using Explainable Deep Learning techniques.

BACKGROUND: Lung cancer has the highest mortality rate in the world, twice as high as the second highest. On the other hand, pathologists are overworked and this is detrimental to the time spent on each patient, diagnostic turnaround time, and their success rate. OBJECTIVE: In this work, we design, implement, and evaluate a diagnostic aid system for non-small cell lung cancer detection, using Deep Learning techniques. METHODS: The classifier developed is based on Artificial Intelligence techniques, obtaining an automatic classification result between healthy, adenocarcinoma and squamous cell carcinoma, given an histopathological image from lung tissue. Moreover, a report module based on Explainable Deep Learning techniques is included and gives the pathologist information about the image's areas used to classify the sample and the confidence of belonging to each class. RESULTS: The results show a system accuracy between 97.11 and 99.69%, depending on the number of classes classified, and a value of the area under ROC curve between 99.77 and 99.94%. CONCLUSIONS: The classification results obtain a substantial improvement according to previous works. Thanks to the given report, the time spent by the pathologist and the diagnostic turnaround time can be reduced.

CAVIAR: a 45k neuron, 5M synapse, 12G connects/s AER hardware sensory-processing- learning-actuating system for high-speed visual object recognition and tracking.

This paper describes CAVIAR, a massively parallel hardware implementation of a spike-based sensing-processing-learning-actuating system inspired by the physiology of the nervous system. CAVIAR uses the asychronous address-event representation (AER) communication framework and was developed in the context of a European Union funded project. It has four custom mixed-signal AER chips, five custom digital AER interface components, 45k neurons (spiking cells), up to 5M synapses, performs 12G synaptic operations per second, and achieves millisecond object recognition and tracking latencies.