Home
Machine Learning (ML) research within medicine and healthcare represents one of the most challenging domains for both engineers and medical specialists. One of the most desired tasks to be accomplished using ML applications is represented by disease detection. A good example of such a task is the detection of genetic abnormalities like Down syndrome, Klinefelter syndrome or Hemophilia. Usually, clinicians are doing chromosome analysis using the karyotype to detect such disorders. The main contribution of the current article consists of introducing a new approach called KaryML Framework, which is extending our previous research: KarySOM: An Unsupervised Learning based Approach for Human Karyotyping using Self-Organizing Maps . Our major goal is to provide a new method for an automated karyotyping system using unsupervised techniques. Additionally, we provide computational methods for chromosome feature extraction and to develop an intelligent system designed to aid clinicians during the karyotyping process.
Keywords: Unsupervised learning, Fuzzy Self-Organizing Maps, Karyotyping, Chromosomal disorders.
- Detection of genetic disorders (e.g. Down syndrome, Klinefelter syndrome) is done with help of karyotypes
- Karyotyping requires a lot of manual work and expert knowledge - takes up to 2 weeks
- Many approaches on karyotyping problem, but there still exists a need for effciency and accuracy
- A new approach on karyotyping process gives us a new branch that have to be covered in order to obtain a better solution for this problem
- Contains 6 different karyotypes and their expected output
- One instance of Down Syndrome
- Chromosome length and area highly correlated as expected
Conducted in 3 steps: PreRuns, KarySOM re-evaluation, Scoring runs
45 different models tested using both classiffers.
- All combinations of the 4 features using Euclidean distance -> 15
- All combinations of the 4 features using Weighted Euclidean distance with multiple weight sets for each -> 30
Selected top performing 16 models (including models defined in KarySOM) for these two steps
Supervised approaches measuring pair number assignment accuracy not pairing accuracy.
In following plot is presented the comparison between our original aproach(KarySOM) and the new one(KaryML Framework).
If this is helpful in your research, cite us:
Casian-Nicolae Marc. A study towards using unsupervised learning in automating Human Karyotyping. In Studia Universitatis Babes-Bolyai, Informatica, 2020