This document discusses artificial intelligence and its applications in radiology. It begins with definitions of artificial intelligence and its subsets of machine learning and deep learning. It then discusses how machine learning and deep learning are being used in medical imaging for tasks like cancer diagnosis and detection of findings in images. The document outlines how large amounts of medical image and patient data are being used to train AI models to perform tasks like segmentation and anomaly detection. It provides examples of startups and projects applying AI to problems in radiology. It concludes by discussing views on AI in radiology, noting that AI can increase radiologist efficiency and consistency if integrated into the workflow, rather than replacing radiologists.
2. ARTIFICIAL INTELLIGENCE
• Branch of computer science devoted to creating systems to perform tasks that
ordinarily require human intelligence.
• Creation of intelligent machines that work and react like humans
Reasoning, planning, learning, language processing, perception and the ability
to move and manipulate objects.
3. Medical imaging
• Medical diagnosis
• Autonomous vehicles (drones and self-driving cars)
• Search engines (such as Google search), online assistants (such as Siri)
• Creating art ,mathematical theorems
• Image recognition in photographs
• Predicting flight delays
5. MACHINE LEARNING
• Subfield of artificial intelligence.
• Expert humans discern and encode features that appear distinctive in the data
(Feature engineering)
• Algorithms are trained to perform tasks by learning patterns from data rather
than by explicit programming
• Machine learning uses algorithms to learn from data and make informed
decision on what it has learned.
7. • Set of available inputs and desired outputs.
• Common inputs in radiology are image data and report text.
8. • Human versus computer
vision.
Have to apply machine learning
algorithms to distinguish images on
the basis of these features
9. DEEP LEARNING
• Subfield of machine learning.
• No feature engineering is used.
• Instead, the algorithm learns on its own the best features to classify the
provided data.
• Structure algorithms in layers to create artificial neural network and can
learn and make decision.
10. Convolutional neural network
• Convolutions are mathematic transformations (similar to a basic filter in a
photograph editing application) that are applied to pixel data.
11.
12. ARTIFICIAL NEURAL NETWORK
• Artificial neural network is a biologically inspired network of artificial neurons
configured to perform specific task.
• Neural network acquires knowledge through learning.
• Contains large number of artificial neurons arranged in series
• Based on a neural network structure loosely inspired by the human brain –
Convolutional neural network
13. ARTIFICAL NEURAL NETWORK (ANN)
• ANNs are clusters of interconnected nodes, like brain neurons.
• Feed those blocks into multiple (deep) processing layers, which act as filters,
and then feed data onto further layers with other kinds of filters.
16. • Oncology - assisting clinical decision making related to the diagnosis and risk
stratification of different cancers.
• non-small-cell lung cancer (NSCLC) used radiomics to predict distant
metastasis in lung – adenocarcinoma and tumour histological subtypes as well
as disease recurrence, somatic mutations, gene-expression profile and overall
survival.
17.
18. To train a model, we need data.
• Massive amounts of digital data now available to train algorithms and
modern, powerful computational hardware
• Radiomics: Medical study that aims to extract large amount of
quantitative features from medical images using data-
characterization algorithms
19. • Radiographic images, coupled with data on clinical outcomes - rapid
expansion of radiomics as a field of medical research
20.
21. DEEP LEARNING SYSTEM
• DEEP MIND GOOGLE
• Struck a deal with NHS
• 700 scans of head and neck cancer to be given to system so that areas to be
treated and avoided during radiotherapy can be delineated.
• 1 million eye scans with information to be given to teach it to recognize eye
illness.
22. • Massachusetts General Hospital (MGH) and Brigham and Women’s Hospital
(BWH) spent more than $1 billion on data collection infrastructures.
• Recently, these institutions started a Center for Clinical Data Science to
produce new clinical AI applications, trained with these data,
23. RECENT USES
• Efforts to enable radiologists to utilize AI as part of their normal PACS workflow.
• Triaging studies that need urgent review by radiologists.
• Facilitating the communication of urgent results.
24. Detecting
• Intracranial hemorrhage
• Chest Xray diagnosis
• Classifying liver lesions on MRI scans and explaining the findings,
• Characterizing pulmonary nodules on CT
• Helping to avoid unnecessary thyroid nodule biopsies.
25.
26.
27. AI RADIOLOGY START UPS
• Zebra technologies
• Using million high quality images to
develop DL engine to automatically detect
various medical findings.
• Automatic detection of liver, lung CVS
and bone diseases.
• Deployed at more than 50 hospitals
globally
28. AI RADIOLOGY START UPS
AIDENCE
• Product provides fully automated
analysis and reporting of pulmonary
nodules.
• Accuracy levels equal and even
surpassing human capabilities.
Artery’s
FDA clearance for product that provides
automated, editable ventricle
segmentations based on cardiac MRI
images as accurate as performed
manually.
29. AI RADIOLOGY STARTUPS
• Butterfly Network
• Hand held probe connected to
iPhone. Starts 2000 dollars
• DL system will identify
characteristics in image and
make diagnosis.
30. INDIAN START UPS
QURE.AI
Deep learning based diagnosis
of tuberculosis on chest x ray
and intracerebral hemorrhage
on CT
Processed more than 1.5
million x-rays with accuracy
level near to humans.
Deployed across 4 clinics in
mumbai.
32. Artificial intelligence
• Fear has been AI would begin to chip away at jobs.
• While that concern isn’t coming true as yet.
• Radiologists are being urged to accept and incorporate AI into their
interpretations.
33. Artificial intelligence
• Primary driver - desire for greater efficacy and efficiency in clinical care.
• Studies report that, in some cases, an average radiologist must interpret one image
every 3–4 seconds in an 8-hour workday to meet workload demands
• Involves visual perception as well as independent knowledge, errors are inevitable
• Integrated AI component within the imaging workflow - increase efficiency, reduce
errors and achieve objectives
• providing trained radiologists with pre-screened images and identified features.
34. • Improves their consistency and quality and potentially lowers operating costs.
• Scientists have shown that few pixels from other image can drastically alter
results.
• Artificial intelligence won’t necessarily replace radiologists, but it will replace
radiologists who don’t use artificial intelligence in future.
More amount of data is fed to the machine – it will learn and make a more acuurate diagnosis
take various attributes and program it to an algorithm based on featues such as edges, gradients, and textures. // Statistical analysis of the presence of these features in a given image can then be used to classify or interpret the image.//
And make associations and form output statistical probabilities, also known as nodes.
A human expert easily classifies this image as an image of the right kidney. // computer “sees” a matrix of numbers representing pixel brightness. Computer vision typically involves computing the presence of numerical patterns (features) in this matrix, then applying machine learning algorithms to distinguish images on the basis of these features.
Complex algorithms , which use multiple lays of processing
Gabor filters related to texture -
The imaging featues are known as evidence .. Many such evidence s are evaluated in layers which gets more complex.. and are integrated-- outputs a decision signal based on a weighted sum of evidences, and an activation function, which integrates signals from previous neurons. Hundreds of these basic computing units are assembled together to build an artificial neural network computing device. \
Fig. 3 |. Artificial intelligence impact areas within oncology imaging.
This schematic outlines the various tasks within radiology where artificial intelligence (AI) implementation is likely to have a large impact. a | The workflow comprises the following steps: preprocessing of images after acquisition, image-based clinical tasks (which usually involve the quantification of features either using engineered features with traditional machine learning or deep learning), reporting results through the generation of textual radiology reports and, finally, the integration of patient information from multiple data sources. b | AI is expected to impact image-based clinical tasks, including the detection of abnormalities; the characterization of objects in images using segmentation, diagnosis and staging; and the monitoring of objects for diagnosis and assessment of treatment response. TNM, tumour–node–metastasis.
Deep learning is a subset of machine learning that is
Radiomics studies have incorporated deep learning techniques to learn feature representations automatically from example images
Brain Tumor Image Segmentation (BRATS)
Much new radiology conferences have been focusing on artificial intellingenc eas the
Deep learning
In latest conferences have been urged .. Keep utptodated