Convolutional neural networks in dermatology

What is a convolutional neural network?

A convolutional neural network (CNN) is a type of artificial intelligence deep neural network used in image processing [1]. The network takes an input image and using a catalog of existing images produces an output that matches the input image. a neural network is based on similarities of interconnected biological neurons and is structured in such a way to learn and improve its performance, based on how many images the CNN 'sees' and the number of convolutions (combining inputs of two images, a new one and an existing cataloged image, to create a third output) that generates CNN.

CNNs are considered a beneficial new tool for dermatologists to help better diagnose lesions. The work a CNN does to produce a diagnostic result from an image is similar to how a dermatologist uses their training and knowledge: diagnosing lesions by a dermatologist usually involves an input image (of a cutaneous injury) feed through a processing network (the skills and knowledge of the dermatologist who analyzes it and synthesizes the available information) to generate a “class” (or diagnosis) or a “class probability” (differential diagnosis) [2].

The visual nature of dermatology it lends itself well to digital imaging of injuries, and CNNs have enormous potential to change practice. It is a multifaceted means of analyzing data that involves complex mathematics and requires massive computational power to combine biology, mathematics, and computer science.

Who uses convolutional neural networks?

CNNs have been used in military and civil applications, including unmanned aerial vehicles, the technology sector, and commerce. [3]. They are found in everyday applications, such as social media platforms that automatically recognize faces, self-adhesive photo galleries, and shopping websites that present suggestions based on your Internet browsing habits.

In the medical field, researchers have been using CNN to diagnose diabetic eye disease, arrhythmiasand skin cancers [3–5].

Tell me more about convolutional neural networks

The basis of a CNN is a computational method that is capable of differentiating between different classes of images based on unique features that can be reliably used to identify the class of image, such as edges and curves. This base is extended with more abstract features that are added through a series of convolutional pools and output layers.

Step 1: Convolutional Layers

An image is fed into a computer and processed as different arrangements of pixels (dots), based on their color. This process converts sections of the image according to a particular filter [2]. Filters often start by analyzing simple features such as straight lines, diagonal lines, curved lines, or points. Every time a filter is passed over the original image, a new, smaller version of the original photograph is created. Positive filter matches are assigned a positive value, and unmatched areas are assigned a value less than 1. This results in a jumbled image; for example, a straight line filter passed over an image of a acral nevus with a parallel sulcus pattern on dermoscopy will show a strong positive convolution image. This step can be repeated with other features to achieve a more precise result (ie, a diagnosis or differential diagnosis).

Step 2: Grouping Layer

If the resulting image is large, the neural network layers may need to “group” between subsequent convolutions, looking at an area of interest in the image and removing parameters around that area [2]. There are different types of pooling, but the most common type used is the maximum pooling method (see figure below).

There is often an additional normalization step, which is a common technique to improve the performance and stability of a neural network. This works by standardizing the inputs analyzed by the neural network, to ensure that each input has roughly the same scale. Therefore, the neural network will not assign undue importance to one input filter over another simple one due to a difference in scale. This process greatly increases the learning rate of a neural network.

Step 3: output layer

To generate a differential diagnosis output for a suspicious lesion, the neutral network needs to apply a fully connected layer based on all the layers it has previously processed. This is similar to a dermatologist synthesizing the different clues into a tentative diagnosis with a set of differentials.

The CNN can now be trained using additional functions to improve accuracy and "teach" itself to identify new lesions (such as backpropagation, which teaches the network when it selects an incorrect result to change the weight assigned to features when select an output class) [1].

What are the benefits of convolutional neural networks?

The benefits of CNNs in diagnosing skin lesions include accuracy, speed, and low cost.

• The clinical diagnostic accuracy for melanoma it is dependent on the experience and training of the medical examiner; CNNs have been able to perform as well as board-certified dermatologists in limited circumstances, and their accuracy will continue to improve in the future [6,7].
• Currently, CNNs take seconds to minutes to reach a diagnosis when confronted with an image of a skin lesion. Inputs, algorithms and outputs can be done outside office hours and can be accessed by anyone with internet access. Compare this short time to the wait and travel times associated with a dermatologist appointment, which is often several months in the future or more.
• Algorithms can be adaptive and can learn to add new images over time.
• CNNs are predicted to be able to diagnose lesions for a fraction of the cost of a visit to a dermatologist.

What are the disadvantages of convolutional neural networks?

Warnings about the use of CNN include unrealistic expectations of patients and healthcare professionals, security and privacy concerns, and legal medical liability.

• There is a great deal of excitement surrounding CNN technology, but the benefits will take time to materialize. Massive amounts of data and information are required to “train” CNNs. Human beings are needed to choose which injuries should be screened and reviewed by CNN, and these participating health professionals also need training.
• CNNs and any tools that offer diagnostic support will need to be officially approved as medical devices, and then re-approved as their algorithms expand [8].
• CNNs will likely be entirely online, using cloud-based storage, and will need to have excellent cybersecurity systems in place to ensure backup in case of database or server failure, and authentication processes to prevent unauthorized access. (Encryption and secure transfer protocols are required to store personal health data, and research should only use anonymous data.)
• Healthcare professionals using CNN should understand that performance with one data set is not necessarily applicable to another; there will be incorrect diagnoses, including false positives (overdiagnosis of benign injuries like evil one) and false negatives (eg, missed diagnoses of Cancer)
• The medicolegal responsibility for the health professional who depends on the CNNs requires clarification, since there is no notable precedence. Can a computer algorithm be held responsible for an incorrect diagnosis or a missed diagnosis?