2) Deep Learning

What is Machine Learning?

Machine learning is a type of artificial intelligence (AI) where it’s intelligence is exhibited by software applications in order to take actions or make predictions without having to be programmed. Machine learning algorithms can be classified under two categories, either as supervised or unsupervised. As the name suggests, supervised algorithms in a system require both input and desired output data output are provided. Both of these parameters are provided to the system in order to provide labelled classification to provide a learning basis for future data processing. On the other hand, unsupervised learning is training without guidance and uses information that is classified nor labeled. [1]

[5]

Deep Learning Black Box

Sebastian Thrun from Stanford University was successfully able to demonstrate that a deep learning algorithm was able to correctly diagnose a potentially danger mole. This algorithm was able to outperform 21 dermatologists in identifying which moles were cancerous. Although the predictions was correct, people were not able to tell which features of a mole the deep learning algorithm were used or how they were used to classify whether or not it was cancerous or benign. In the case of black box medicine, doctors do not know how it functions as they are very complex algorithms.

The FDA has certified numerous image analysis applications that rely on deep learning algorithms and noted that the companies are allowed to keep the details for their algorithms confidential. One example of this is a software called “Deep Ventricle” that analyzes MRI images of the interior contours of the heart's chambers and calculates the volume of blood a patient's heart can hold and pump and can do this in under 30 seconds.

The worry that doctors will be replaced with these sort of Deep learning black boxes is absolutely out of the question as tests will still need to be analyzed by doctors. [2]

[6]

Application Areas [3]

Machine learning is used in the medical imaging field and facilitates several services like computer-aided diagnosis, image segmentation, image registration, image annotation, and image-guided therapy. There are three main applications of deep learning in healthcare.

[1]

Faster Diagnosis

Doctors use medical imaging techniques such as CT scans, MRIs, and X-rays for diagnosing conditions ranging a wide spectrum of diseases. The analysis of these images can sometimes be difficult and time consuming. Also doctors can sometimes not see conditions

Genomics For Personalized Medicine

Genomics involves accumulating patient data so that they can study the genetic factors of mutations and lead to disease. It can also help in personalized medicine, with treatments that are customized according to patient’s genomic makeup

Computer Aided Diagnosis

This is a computerized procedure that provides a second opinion for the interpretation of a medical image. The introduction of CAdx, there has been promising results on various medical applications such as the diagnosis and treatment of Alzheimer's disease, organ segmentations amongst others.

Current Deep Learning Medical Applications [4]

Deep Learning methods are still relatively new, but it’s implementations are still quite common.

Tumor Detection

Melanoma is the deadliest form of skin cancer which if diagnosed early can be curable. Proper treatment and early detection can give a 5 year survival rate of up to 98%. This implementation of deep learning is the most promising near term application of deep learning on humans.

Tracking Tumor Development

Deep learning algorithms can be used for track tumor development using medical images. The algorithm will generate tumor probability heatmaps that show overlapping tissues patches classified for tumor probability. Such images provide informative data on different tumor features such as shape, area, density, and location.

Blood Flow Quantification and Visualization

MRI gives a good visualization for blood flow in blood vessels without having to use toxic contrast agents. A deep learning software called Arterys have made it possible for cardiac assessments to occur in a fraction of the time in comparison to conventional cardiac MRI systems.

Medical Interpretation

Usually medical imaging tests reply on both the image and interpretation of the radiologist of the images. A deep learning algorithms can be trained to detect the presence or absence of the diseases in the medical images, helping doctors to come up with better interpretations.

Diabetic Retinopathy

Diabetic Retinopathy is the most severe complication of diabetes in the world which causes blindness. If detected early, it can be treated without side effects. The deep learning algorithm shown in the study is trained on a neural network used to compute the severity on images.