今日吃瓜 researchers find a way to use convolutional neural networks to identify cancer in tissue samples, which could speed up diagnosis and improve outcomes in patients with colorectal cancer.

Colorectal cancer is the third most common cancer among men and women worldwide, and the second most common cause of cancer-related mortality. Most colorectal cancers are due to old age and lifestyle factors, with only a small number of cases due to underlying genetic disorders. It typically starts as a benign tumor, such as a polyp, which over time becomes cancerous. Like all forms of cancer, early diagnosis and differentiation of the tumor are crucial for a patient鈥檚 survival and wellbeing.

Colorectal cancer may be diagnosed by obtaining a sample of the colon and using histopathology 鈥� the study of changes in tissues caused by diseases 鈥� to determine the characteristics of the tumour tissue at the microscopic level.

Histology is the study of the microanatomy of cells, tissues and organs as seen through a microscope. The structure of each tissue in the body is directly related to its function and diseases affect tissues in distinctive ways. Studying the histology of a tissue can be very useful in making a diagnosis and determining the severity and progress of a condition.

Because of the great variety of tests that are available, and the high level of skill needed to carry out and interpret them, researchers are beginning to turn to computational pathology and artificial intelligence techniques to identify in tissue samples diseases like cancer.

Dr. Sajid Javed, Assistant Professor, and Dr. Naoufel Werghi, Associate Professor of Electrical Engineering and Computer Science, have collaborated with researchers from around the world to develop algorithms to identify samples of colorectal cancer tissue. A paper based on this research has been published in.

鈥淐omputational pathology is a fast-growing research area in cancer diagnosis and can play an instrumental role in helping medical professionals detect and classify tumors,鈥� said Dr. Javed.

Cancer histology reveals underlying molecular processes and disease progression and contains rich phenotypic information that is predictive of patient outcomes.

The phenotype is the set of observable characteristics or traits of an organism or a tissue. Image-based phenotyping aims to develop the computer vision techniques and tools needed to recover quantitative data from a wide range of images. But phenotyping presents challenging problems, particularly in images from colorectal cancer tissues.

Aided by advances in slide scanning microscopes and computing, convolutional neural networks (CNNs) have emerged as an important image analysis tool. CNNs use a network of interconnected layers of filters that highlight important patterns in the images and can continue to learn from previous results.

鈥淢anual examination of tissue samples is time-consuming, highly subjective, and often affected by the observer,鈥� explained Dr. Javed. 鈥淢eanwhile, algorithms analyzing digitized Whole Slide Images (WSIs) can examine hundreds of thousands of cells and billions of pixels to differentiate seven distinct tissue phenotypes.鈥�

Deep learning methods require large amounts of annotated histology data for training, which may be tedious to obtain. Additionally, while these methods may be effective in determining tumor tissue, the tissues in colorectal cancer also contain a rich mix of several other types of tissue, including smooth muscle, inflammatory, necrotic, and benign tissue. Any algorithm must be taught to distinguish between these tissue types to be effective.

Texture analysis is a commonly used approach for tissue phenotyping, where texture features are computed to train classifiers, which are then used to predict distinct tissue types.

鈥淭exture analysis may be attractive due to its simplicity but it does not fully capture the biological diversity of tissue components,鈥� explained Dr. Javed.

鈥淩ecent methods have proposed integrating cellular connectivity features, which are used as a proxy to cellular interaction features. The notion of cellular connectivity features is based on the fact that spatially adjacent cells have a higher probability of receiving inter-cellular signals from each other than from cells that are farther away. It has also been shown that inter-cellular signals between various types of cells can influence the progression of cancer. However, a dynamic network of tumor growth cannot be adequately modelled by a single type of interaction. Our technique uses a multiplex network model to represent the intricate relationships between cell populations. We propose four different types of cellular networks integrating a variety of features representing tissue characteristics at different levels.鈥�

In the researchers鈥� model, cells from a WSI are detected and classified into five distinct categories using a deep neural network. Then, four different types of cellular interaction features are computed and used to construct a four-layer multiplex graph. Since each slide contains thousands of cells, the slides are segmented into tiles or patches, which helps the algorithm determine the distribution of different types of tissues across the cells.

鈥淭here are many directions in which this work can be further extended,鈥� added Dr. Javed. 鈥淔urther cellular types such as blood cells could improve performance and also reveal more micro-level tissue communities. Additionally, our framework could be adapted to WSIs of different types of cancer. Of course, in clinical practice, our work can help medical practitioners understand the contents of the WSI and make more accurate and timely diagnoses.鈥�

This work was funded by the Khalifa University of Science and Technology and the UK Medical Research Council. The collaborators were also supported by the PathLAKE digital pathology consortium, which is funded from the Data to Early Diagnosis and Precision Medicine strand of the UK government鈥檚 Industrial Strategy Challenge Fund, managed and delivered by UK Research and Innovation.听

Jade Sterling
Science Writer
5 July 2021

The post Leveraging AI to Detect Colorectal Cancer appeared first on 今日吃瓜.

The original computers were designed around a human brain model. Since then, developments in artificial intelligence and computer science continue to take inspiration from the brain.

Read Arabic story .

The human brain has always been under study for inspiration of computing systems. Although there鈥檚 a very long way to go until we can achieve a computing system that matches the efficiency of the human brain for cognitive tasks, several brain-inspired computing paradigms are being researched. Convolutional neural networks are a widely used machine learning approach for AI-related applications due to their significant performance relative to rules-based or symbolic approaches. Nonetheless, for many tasks machine learning requires vast amounts of data and training to converge to an acceptable level of performance.

A PhD student from Khalifa University, Eman Hasan, is investigating another AI computation methodology called 鈥榟yperdimensional computing鈥�, which can possibly take AI systems a step closer toward human-like cognition. The work is supervised by Dr. Baker Mohammad, Associate Professor and Director of the System-on-Chip Lab (SOCL), and Dr. Yasmin Halawani, Postdoctoral Fellow.

Hasan鈥檚 work, which was published recently in the journal, analyses different models of hyperdimensional computing and highlights the advantages of this computing paradigm. Hyperdimensional computing, or HDC, is a relatively new paradigm for computing using large vectors (like 10000 bits each) and is inspired by patterns of neural activity in the human brain. The means by which can allow AI-based computing systems to retain memory can reduce their computing and power demands.

HDC vectors, by nature, are also extremely robust against noise, much like the human鈥檚 central nervous system. Intelligence requires detecting, storing, binding and unbinding noisy patterns, and HDC is well-suited to handling noisy patterns. Inspired by an abstract representation of neuronal circuits in the human brain, developing an HDC architecture involves encoding, training, and comparison stages.

The human brain is excellent at recognizing patterns and using those patterns to infer information about other things. For example, humans generally understand that just because a chair is missing a leg, that doesn鈥檛 mean it鈥檚 no longer a chair. An AI system may look at this three-legged chair and decide it is a completely new object that needs a new classification. HDC vectors, however, offer some margin for error. With HDC, recognizing certain features will generate a vector that is similar enough to a chair that the computer can infer the object is a chair from its memory of what a chair looks like. Hence, the three-legged chair will remain a chair in hyperdimensional computing while in traditional object recognition this is a difficult task.

鈥淚n a HD vector, we can represent data holistically, meaning that the value of an object is distributed among many data points,鈥� explained Hasan. 鈥淭herefore, we can reconstruct the vector鈥檚 meaning as long as we have 60% of its content.鈥澨�

The structure of the vectors leads to one of the strongest advantages of the HDC approach, which is that it can tolerate errors and therefore is a great option for approximate computing applications. This arises from the representation of the hyper vectors, where a bit value is independent of its location in the bit sequence.

HDC is also powerful in that it is memory-centric, which makes it capable of performing complex calculations while requiring less computing power. This type of computing is particularly useful for 鈥榚dge鈥� computing, which refers to computing that鈥檚 done at or near the source of data. In a growing number of devices, including in autonomous vehicles, computations must be carried out immediately and at the point of the data collection, instead of relying on computing done in the cloud at a data center.

听

鈥淗yperdimensional computing is a promising model for edge devices as it does not include the computationally demanding training step found in the widely used convolutional neural network,鈥� explained Hasan. 鈥淗owever, hyperdimensional computing comes with its own challenges as encoding alone takes about 80 percent of the execution time of its training and some encoding algorithms result in the encoded data growing to twenty times its original size.鈥�

Hasan studied the HDC paradigm and its main algorithms in one-dimensional and two-dimensional applications. Research has shown that HDC outperforms digital neural networks in one dimensional data set applications, such as speech recognition, but the complexity increases once it is expanded to 2D applications.

鈥淗DC has shown promising results for one dimensional applications, using less power, and with lower latency than state-of-the-art simple deep neural networks,鈥� explained Hasan. 鈥淏ut in 2D applications, convolutional neural networks still achieve higher classification accuracy, but at the expense of more computations.鈥�

Hasan concluded that HDC is still considered a new paradigm and faces challenges requiring further analysis.听

Jade Sterling
Science Writer
29 April 2021

The post A Step Closer to Brain Like AI with Hyperdimensional Computing appeared first on 今日吃瓜.