How Deep Learning-Based Image Analysis Software performs Medical DiagnosticEstimated Reading Time: just 4 min

Artificial Intelligence (AI) is the intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans and other living beings.

For a machine acting like a human means being able to perform 5 macro-functions:

  • Perception (recognizing and interpreting reality)
  • Reasoning (drawing new conclusions)
  • Planning (detecting patterns and adapting to new environments)
  • Knowledge Representation (extract and store knowledge)
  • Natural Language Processing (communicate successfully in a human language)

AI research started in 1956 after a workshop at Dartmouth College and its fathers where:

  • Allen Newell and Herbert Simon from CMU
  • John McCarthy and Marvin Minsky from MIT
  • Arthur Samuel from IBM

Dartmouth College group focused on Restricted AI that solves only one of the 5 acting like a human macro-functions at a time.

Recently, a huge breakthrough came with the birth of Machine Learning.

Machine Learning is a field of artificial intelligence that uses statistical techniques to give computer systems the ability to learn from data, without being explicitly programmed.

This step ahead allowed systems to solve problems that involve:

  • Supervised Learning
  • Clustering
  • Dimensionality Reduction
  • Structured Prediction
  • Anomaly Detection

Another breakthrough came with Deep Learning.

Deep Learning is a subset of Machine Learning that, using architectures based on Neural Networks, allows machines to learn a large variety of activities as humans do.

This step ahead allowed systems to solve problems that involve:

  • Supervised/Semi-Supervised/Unsupervised Learning
  • Image/Audio Recognition
  • Natural Language Processing

Nowadays, there are 3 main problems that can be solved with AI

  1. Regression Problems: this could be applied whenever the output of a subset of inputs is a continuous function
  2. Classification Problems: this could be applied whenever the output of a subset of inputs is a discrete function
  3. Clustering Problems: this could be applied whenever the output of a subset of inputs is not predictable/known

The 3 problems are solved by using sequences of algorithms called Neural Networks.

Usually, every Neural Network performs 4 activities:

  1. Data Collection
  2. Data Cleaning
  3. Patterns Identification
  4. Prediction Making

One of the most interesting applications of Artificial Intelligence is performing Medical Diagnostic.

When diagnosing a disease, the stage of the illness is one of the key factors.

In fact, the earlier a diagnosis is made and the greater are the chances to cure it.

Sometimes there are symptoms of a disease that a human cannot see.

Let’s think, for instance, about looking at an x-ray of lungs.

If cancer is just beginning, a doctor may see it by using a microscope but the operation will require a very long time and the chance of mistakes would be high.

The same operation can be done by using a deep learning-based image analysis software.

This kind of software can use Artificial Intelligence in order to perform:

  • Defect detection
  • Material classification
  • Deformed part location

Let’s now think about an electrocardiogram graph.

If there is a small variation of the curves suggesting there is a risk for a stroke to happen in the next 6 months, a doctor may not see it and the chance of the symptom to pass unnoticed is high.

The same operation can be done by using a deep learning-based clustering software.

This kind of software can use Artificial Intelligence in order to perform clustering analysis of tests output. Each cluster can be then related to a diagnosis.

Such tools are super-human since they can analyse a huge number of images very quickly with:

  • A detail that is greater than a human eye looking into a microscope
  • Accuracy and consistency that are greater than a human who gets tired or may be distracted

Most of all, these applications are real already and some examples are:

  • To determine whether something is wrong with a patient’s heart, a cardiologist will assess the timing of their heartbeat in scans. In the UK, the correct diagnosis rate is estimated to be 80%. To improve this KPI, the John Radcliffe Hospital in Oxford, England, developed an AI software called Ultromics. After being trained using the heart scans of 1,000 patients, the software had a correct diagnosis rate of 90% that will improve with time.
  • A study of the American Cancer Society says that the 5-year survival rate for lung cancer is about 90% when the diagnosis is made in the early stages while dropping to 10% in the later ones. Pulmonary function tests provide an extensive amount of data for a diagnosis, but it can be difficult for a human eye to catch all the patterns and insights. Therefore, the Laboratory for Respiratory Diseases of the Catholic University of Leuven, Belgium, developed an Artificial Intelligence system to manage these data. After being trained using the tests of 1,430 patients from 33 Belgian hospitals, the algorithm proved to be more accurate in twice as many cases as a diagnosis by pulmonologists. The AI system is now incorporated into the real clinical practice at the University Hospital of Leuven.
  • In the US, early detection of skin cancer is correlated with a survival rate of 95% but quickly decreases with later stages, hitting 15%. Stanford University researchers have trained an algorithm to diagnose skin cancer using deep learning, specifically deep convolutional neural networks (CNNs). The algorithm was trained to detect skin cancer or melanoma using 130,000 images of skin lesions. The algorithm was tested against 21 board-certified dermatologists. The results showed that the Artificial Intelligence algorithm had the same ability as the 21 doctors in determining the best course of action across all images.

These are only some of the many examples about Artificial Intelligence applied to Medical Diagnostic but, at the same time, they are representative of the power behind them.

In fact, the vision is to collect all the data related to patients treatments occurred all around the world and to use them to train such deep learning-based systems in order to be able to use machines to diagnose diseases accurately and very early while focusing the humans in deep diving any insight received and in curing the patients at their best.

Nicola Zaffonato Administrator

Business Strategy | Product Marketing | Executive Master eCommerce Management | Business Innovation Master | MSc

I am driven by my personal growth and of people/contexts that surround me.

I followed a professional path in Valentino Fashion Group and Luxottica during which, thanks to the ability to understand different businesses and interests, I was able to succeed in Operations, Merchandising and Retail.

These organizations have exploited my ability to mediate and translate needs/constraints into practice, assigning me to Project Management roles.

Luxottica relied on my ability to analyze, to anticipate things and to imagine/implement solutions by appointing me in Supply Chain Management department and assigning me to the Product Management of IoT solutions for Anti-counterfeiting and Retail digitalization.

During this professional path, I also developed my leadership by managing teams to build Processes, Organizations, Systems and Governance Tools.

follow me
Sharing is Caring!
search previous next tag category expand menu location phone mail time cart zoom edit close