There are two types of ML applications: one is narrow in nature, and the second is broad in nature. Narrow ML deals with creating programs algorithms and robotics software that caters to a narrow focused set of activities or skill set. Here, the narrow means that the area of application is a specialized skill. It relates to an expert and its purpose is to emulate and exceed the human expert in their specialized skill. Narrow ML works best when it has an expert to learn from and to copy. An example of narrow ML robots would be the robotic arms belt for doing heart operations, such as removing blood clots from arteries. This example is of a robot that requires assistance from a human being in order to carry out its operation. Refer to the figure shown below which shows a comparison between narrow and broad ML.
In the figure shown above we can clearly see that narrow ML concentrates on things like healthcare, finance, and retail. In comparison, broad ML is about building a humanoid, giving it cognitive capabilities in artificial intelligence (AI) and the ability to emulate physical characteristics of human being.
Now let us look at the broad ML application. Here, we are talking about creating programs algorithms and Robotics software that caters to generalized skill as opposed to specialized skill. It emulates general human behavior, and the purpose is to prove robotic capability equal to that of a human being. A recent example of such broad application of ML is the robot named Sophia that has gained citizenship in the Kingdom of Saudi Arabia due to its proven ability to emulate human conversation.
As the technology advances we will see more robots being developed on broad ML applications. However, the current trend in the healthcare industry is to adopt robotics and its applications in a narrow way and to help emulate or replace experts in diagnosis of disease research of new drugs and other such areas.
The current state of the healthcare industry has the two opposing forces: one that is the traditional healthcare institution that is generally comprised of wellness clinics, doctor clinics, and hospitals. Another new set of institutions that are coming up are based on robotics ML AI.
The traditional healthcare system derives its values from empathy, human touch, and healing through the doctor. As opposed to this, there is another set of institutions that are coming up rapidly. The values that these institutions bring forward are those of efficiency and accuracy of healthcare operations, better management of resources, and minimal human touch to avoid spread of communicable diseases.
Both the systems target giving better care to the patient. In the traditional view the doctor is irreplacable and is the center of the healthcare institution. However, the new and modern view is that the doctor has a limited capacity of analysis and cannot analyze the big picture—hence, such machine algorithms and robots, which can do a better job. The institutions based on robotic ML and AI are trying to make headway into replacing the traditional healthcare system by targeting narrow ML applications first. Here the attempt is not to replace the doctor as a whole but to replace or emulate and then replace certain specialized functions of a doctor or healthcare professional.
One example of ML being used for narrow healthcare tasks comes from Siemens Company from the division healthineers. They have computer vision imaging on computer tomography and look at what the brain wiring looks like through a MRI scan. They have brain anatomy machines known as Tesla machines. The other application of ML by the same company is the CT scanner, which is used for parametric imaging or molecular imaging, and healthcare workers have applied it to show whether a tumor is benign or malignant.
The ML algorithm has now enabled positioning of patients as quickly as possible to get better images. They have also developed deep learning algorithms for reading chest X-rays and to detect abnormality in the X-ray machine. This is an attempt to replace the specialized role of radiologist with numerous hours of expertise with all X-rays that are thrown before them, including an MRI and CT scan. On the same line, Siemens has developed an MRI image fingerprinting technique using deep learning to emulate what a radiologist does. It is also a pioneer in the field of lab robotic automation, using an electromagnetic levitation technique, which is used in speed trains around the world.
Niramai is another example of an organization using ML applications to develop a solution for overcoming a social barrier in an innovative way. They developed solutions for identification of
breast cancer in women.
In a country like India, where traditional beliefs are prevalent in the rural regions, the major hindrance to detecting breast cancer is that the traditional system requires a doctor to touch the patient’s breast to detect a lump that may become cancerous. The major method used even today is for the doctor to feel and use his/her hands to see if there is a presence of a lump in the region of the body. To overcome this drawback, Niramai looked at how technology could be used to help diagnose breast cancer without using touch or invasive procedures or applying pressure through mammography, which is painful. So they looked at a solution by using high-resolution,full-sensing thermal image with ML and use images to detect prevalence of cancer. By using a high-resolution thermal sensing device and artificial intelligence and ML, they are able to develop API, which is non-invasive, does not require any test, and does not cause any pain to the patient. They require permission to take off the patient’s clothes while the machine detects for the prevalence of cancer and whether it is malignant or benign, which matches any mammography test done manually. Over time the algorithm will learn and improve by itself. Such innovative use of technology that focuses on overcoming social issues in healthcare are going to be adopted faster in countries where the population is high and there are social stigmas against medical help that are preventing it from spreading as a method of cure with the common population.
Healthcare deals with human life, and when we apply ML we need to have a gradual and careful way of adopting technology, as a lot is at stake here. Robust algorithms with prediction models with higher accuracy levels are required. This can be changed from a very simple example where we build a prediction model that predicts a particular type of cancer with an accuracy of 95 percent. In this case the prediction model will predict accurately for 25 patients and predict incorrectly for the other 5 patients. So the incorrectly predicted patients will still think they do not have cancer. This is the reason why application of ML in healthcare requires more testing before a model is deployed in
production.
Here I am ending today's post and will be back with another post focusing on ML. Till then keep exploring and learning.
In the figure shown above we can clearly see that narrow ML concentrates on things like healthcare, finance, and retail. In comparison, broad ML is about building a humanoid, giving it cognitive capabilities in artificial intelligence (AI) and the ability to emulate physical characteristics of human being.
Now let us look at the broad ML application. Here, we are talking about creating programs algorithms and Robotics software that caters to generalized skill as opposed to specialized skill. It emulates general human behavior, and the purpose is to prove robotic capability equal to that of a human being. A recent example of such broad application of ML is the robot named Sophia that has gained citizenship in the Kingdom of Saudi Arabia due to its proven ability to emulate human conversation.
As the technology advances we will see more robots being developed on broad ML applications. However, the current trend in the healthcare industry is to adopt robotics and its applications in a narrow way and to help emulate or replace experts in diagnosis of disease research of new drugs and other such areas.
The current state of the healthcare industry has the two opposing forces: one that is the traditional healthcare institution that is generally comprised of wellness clinics, doctor clinics, and hospitals. Another new set of institutions that are coming up are based on robotics ML AI.
The traditional healthcare system derives its values from empathy, human touch, and healing through the doctor. As opposed to this, there is another set of institutions that are coming up rapidly. The values that these institutions bring forward are those of efficiency and accuracy of healthcare operations, better management of resources, and minimal human touch to avoid spread of communicable diseases.
Both the systems target giving better care to the patient. In the traditional view the doctor is irreplacable and is the center of the healthcare institution. However, the new and modern view is that the doctor has a limited capacity of analysis and cannot analyze the big picture—hence, such machine algorithms and robots, which can do a better job. The institutions based on robotic ML and AI are trying to make headway into replacing the traditional healthcare system by targeting narrow ML applications first. Here the attempt is not to replace the doctor as a whole but to replace or emulate and then replace certain specialized functions of a doctor or healthcare professional.
One example of ML being used for narrow healthcare tasks comes from Siemens Company from the division healthineers. They have computer vision imaging on computer tomography and look at what the brain wiring looks like through a MRI scan. They have brain anatomy machines known as Tesla machines. The other application of ML by the same company is the CT scanner, which is used for parametric imaging or molecular imaging, and healthcare workers have applied it to show whether a tumor is benign or malignant.
The ML algorithm has now enabled positioning of patients as quickly as possible to get better images. They have also developed deep learning algorithms for reading chest X-rays and to detect abnormality in the X-ray machine. This is an attempt to replace the specialized role of radiologist with numerous hours of expertise with all X-rays that are thrown before them, including an MRI and CT scan. On the same line, Siemens has developed an MRI image fingerprinting technique using deep learning to emulate what a radiologist does. It is also a pioneer in the field of lab robotic automation, using an electromagnetic levitation technique, which is used in speed trains around the world.
Niramai is another example of an organization using ML applications to develop a solution for overcoming a social barrier in an innovative way. They developed solutions for identification of
breast cancer in women.
In a country like India, where traditional beliefs are prevalent in the rural regions, the major hindrance to detecting breast cancer is that the traditional system requires a doctor to touch the patient’s breast to detect a lump that may become cancerous. The major method used even today is for the doctor to feel and use his/her hands to see if there is a presence of a lump in the region of the body. To overcome this drawback, Niramai looked at how technology could be used to help diagnose breast cancer without using touch or invasive procedures or applying pressure through mammography, which is painful. So they looked at a solution by using high-resolution,full-sensing thermal image with ML and use images to detect prevalence of cancer. By using a high-resolution thermal sensing device and artificial intelligence and ML, they are able to develop API, which is non-invasive, does not require any test, and does not cause any pain to the patient. They require permission to take off the patient’s clothes while the machine detects for the prevalence of cancer and whether it is malignant or benign, which matches any mammography test done manually. Over time the algorithm will learn and improve by itself. Such innovative use of technology that focuses on overcoming social issues in healthcare are going to be adopted faster in countries where the population is high and there are social stigmas against medical help that are preventing it from spreading as a method of cure with the common population.
Healthcare deals with human life, and when we apply ML we need to have a gradual and careful way of adopting technology, as a lot is at stake here. Robust algorithms with prediction models with higher accuracy levels are required. This can be changed from a very simple example where we build a prediction model that predicts a particular type of cancer with an accuracy of 95 percent. In this case the prediction model will predict accurately for 25 patients and predict incorrectly for the other 5 patients. So the incorrectly predicted patients will still think they do not have cancer. This is the reason why application of ML in healthcare requires more testing before a model is deployed in
production.
Here I am ending today's post and will be back with another post focusing on ML. Till then keep exploring and learning.
0 comments:
Post a Comment