The World Health Organization (WHO) has reported that cancer burden has risen to touch 18.1 million new cases and that about 9.6 million deaths have resulted from cancer in 2018. That said, what this alarming account of cancer deaths reiterates is the need to rollout prevention measures, wherein early detection and efficient prevention need to be embraced to control this disease.
And in making early detection possible, can technology lend a great support in speeding up the process of detecting cancer, augmenting doctor’s decision-making and making it possible to provide treatments as early as it can be done?
Where early detection and treatment of cancer rise in significance, and with deep learning techniques playing a key role in addressing challenges related to medical image processing, image classification based on convolutional neural network (CNN) offers the right solution in augmenting existing practices, speeding up detection, decreasing wait time for getting results and in increasing interaction between the Doctor and patient for speedy diagnosis and care.
Let us take the case of skin cancer, where the WHO reports 1.04 million cases, and explore how image classification based on CNN speeds up cancer type detection and augments decision making.
How image classification fits in?
Let’s take the case of images corresponding to cancer types, pertaining to various patients. In helping detect the cancer type, image classification comes good in the way it allows creation of thematic maps. Now, images help in detection in the way features in the images can be captured to speed up detection. With each image having different features as when compared to other images, what can be acquired are the different classes pertaining to images. In effect, algorithms can be used for image classification and detecting the class associated with an image.
Through image classification, decisions made by radiologists and doctors with regards to cancer type detection can be automated, which in turn decreases the ‘waiting time’ to know the results and optimizes process time allotment for diagnosis.
What type of image data?
For this exercise pertaining to cancer type detection leveraging image classification, let us consider seven types of skin cancer and consider images pertaining to these seven different types. Images used belong to the 3-dimensional PNG format, and these are images pertaining to the following types.
- Melanocytic nevi
- Melanoma
- Benign keratosis
- Basal cell carcinoma
- Actinic keratoses
- Vascular lesions
- Dermatofibroma
With the images sorted out for the image classification exercise, image processing techniques are used to convert the images to the required format and fed into the CNN (Convolutional Neural Networks).
Why Convolutional Neural Network for image classification?
As ‘features’ remain the core element of this image classification problem, CNN rises as the best choice for solving this classification problem of detecting cancer types. It is the ‘Convolution Feature’ of CNN that makes it stand apart from the other Networks belonging to the Neural Networks. This also makes it achieve outstanding results in terms of grasping information or features in this case, and in facilitating faster training and use of fewer samples.
Take the case of the cancer type images. When an image is fed into CNN model, observations enabled by the neural network include the following:
- Abstract features and shapes that make no sense, where the convolutional neural network observes and learns and goes deeper to unearth vague features
- Molecular level abstractions will be observed and learnt as it goes deeper, as that of Pixel, density and shapes
‘Going deeper’ part with respect to neural networks is about bringing sophistication to human perception – moving from detection of features in pixel to shapes and objects. In this process, CNN connects the lower as well as higher level features, creates the right path for the image to be mapped to the right ‘class’ of image. For model building, CNN rises in relevance for detecting dumped images through classification and identifying the cancer type, from among the seven cases considered in this case.
For instance, a radiologist or a doctor can leverage convolutional neural network model to profit from the speedy detection of cancer types as when images are fed into the CNN model and made to go through several layers. The model then provides the output in determining what type of skin cancer and in inferring if it is a cancerous type.
What are the other value-adds?
The ease-of-detection can be furthered by integrating the CNN model with web interface. In addition to the integration established with web interface, Convolutional Neural Network model built to detect skin cancer can also be integrated with any device, be it IOS or Android or even IoT devices. Through this image classification solution, mundane jobs of Radiologists are automated leading to faster detection of skin cancer types. Infusing AI into the detection process can promote early detection, eliminate wait time and enable deduction within just a few clicks.