AI can help predict Skin Cancer Recurrence

 

Hello! Today we have an entry in the field of medicine.

What is skin cancer recurrence?

An early-stage skin cancer type called melanoma is treatable if diagnosed early enough. Most melanoma deaths in the US occur because of recurrence of the disease which was early stage at the time of diagnosis. However, recurrence is not detected until the cancer starts to spread; called symptomatic metastatic progression. This is where prognostic tools can be helpful for regular surveillance and quick action to stave off mortality and ensure health. Also, identifying high-risk patients can help in determining who should receive special therapies and treatments.

Enter Machine Learning.

A team led by investigators at Massachusetts General Hospital performed two types Machine Learning prediction with nine different models:

  1. Melanoma Recurrence Classification: Where the prediction would be a probability that melanoma recurrence would occur.
  2. Time-To-Event Melanoma Recurrence Risk Prediction: Where the prediction would be time taken for the event of melanoma recurrence to definitely occur.

The data sets used for the training and validation were the MGB cohort and the DFCI cohort. MGB is the Mass General Brigham healthcare system and DFCI is the Dana-Farber Cancer Institute.

A comprehensive array of over 36 demographic, clinical, and histopathologic features were used to build the model.

The machine learning models were validated internally and externally. Stratified five-fold cross-validation was done on the model trained using the MGB cohort for internal validation. For external validation, the model trained on the MGB cohort was validated on the DFCI cohort.

Model performance for the recurrence classification was AUC: 0.812. Model performance for the time-to-event recurrence prediction was time-dependent AUC: 0.820.

The predictive capabilities of these models can be improved by incorporating additional features such as digital histopathology images, genomics data, and novel tumor biomarkers. Nevertheless, the researchers opine that these current models can be deployed clinically for detection of high-risk patients who may benefit from early interventions.

This research advancement can be found in full here.

Key Terms:

  1. Melanoma
  2. Histopathologic
  3. Stratified five-fold cross-validation
  4. AUC: Area Under Curve
  5. Time-Dependent AUC


Comments

Post a Comment

Please let us know what you think!

Popular posts from this blog

Introduction

 Hello Fellow Internet User! How has your day been? If you're like me, you're thinking "Um...good or bad or something in between. It is still a great time to be alive!" Welcome to this blog!  Data Science is everywhere! Because data is everywhere! Data Science is fascinating as it is our very human yearning to know  what all of this data tells us . If not out of curiosity then to find out how those stories can help us in various endeavours. Science and Engineering can also benefit from Data Science, maybe even recursively ;-) This blog is meant to bring together a curated set of advancements on how Data Science has been used to further science and engineering endeavours. Apart from drawing your attention to the substance of these advancements, I shall add my own commentary. Should be a fun ride! Hope you enjoy it! Joie de vivre pour toi! Rahul PS: I am open to comments and suggestions. Please let me know what you think of this blog, its posts or anything else of relev
Read more

Crater Detection

Image
Courtesy: James Stuby (NASA) / Wikimedia Commons A crater is a structure or geological formation that is produced as a result of impact of a smaller body upon a larger heavenly body. Craters are found on all solid heavenly bodies such as Earth, the Moon, Mars, Mercury, Venus etc.  Studying craters is important because they are time capsules to how our Solar System evolved and gives us clues to the past, present and future of the evolution of planetary bodies without humans setting foot on them. Crater densities have been correlated with surface age and crater counts have been used used to estimate the relative age of the lunar surface. There are many other uses to studying craters, so it is important to be able to detect craters quickly in an automated fashion. This research uses a novel method of detecting craters by using multiple co-registered data sources: optical images, digital elevation maps (DEMs) and slope maps of the lunar surface. The method uses a deep learning architecture
Read more

AI Helps find Earth-like Habitable Exo-planets

Image
    (Photo credit: BBC) Are We Alone?  It is a question that humans have had ever since we realized there are other worlds, other planets out there in the cosmos. The implications of the answer to this question are staggering no matter what that answer is. In pursuit of an answer scientists have been searching for planets that could potentially harbor life. Planets that are outside of our solar system are termed Exoplanets. The exploratory pursuit is to find "earth-like" exoplanets because these would have the greatest chance of harboring life as we know it. Determining the habitability of exoplanets is not an easy task. Existing ML methods such as Metric-based quantification (using metrics like surface temperature) or Supervised Learning (based mainly on estimated surface temperature) are not reliable enough.  Astronomers from the Indian Institute of Astrophysics along with astronomers from BITS Pilani and Goa campus designed an anomaly detection method to identify potential
Read more

Contact Me!

Name

Email *

Message *

Search Wikipedia For Anything!

Search results