Data Mining Framework for Predicting Ovarian Cancer
Principal Investigator
Name
Roopashri Shetty
Degrees
MTech
Institution
Manipal Institute of Technology
Position Title
Assistant Professor Senior Scale
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-975
Initial CDAS Request Approval
May 9, 2022
Title
Data Mining Framework for Predicting Ovarian Cancer
Summary
The project is part of my PhD research work being pursued in Manipal Institute of Technology, Manipal (Manipal Academy of Higher Education) in collaboration with Kasturba Medical College, Manipal (Manipal Academy of Higher Education).
The project initially aims to develop a scoring system for accurate prediction of ovarian cancer. Current scoring systems do not factor in a lot of recently observed attributes, resulting in high false positivity rate. Hence, to negate that, excessive testing needs to be done for which considerable amount of data is needed.
For developing the scoring system, the dataset must be prepared and balanced first. Though many data preprocessing techniques exist, they are not suitable for medical data because of the high sensitivity. Therefore, as part of our research work, we also aim to come up with data preprocessing techniques, specific to medical data.
Identifying what features affect presence of ovarian cancer in women is also a challenge which we aim to tackle by identifying the relevant features using various techniques.
The current prediction systems completely ignore borderline tumors though they are low-potential malignant tumors. Our research work also involves identifying borderline cases, as a third class, distinguishing them from benign and malignant tumors.
Project later aims to classify ovarian cancer into types like epithelial tumors, metastatic tumor and germ cell carcinoma tumors. Results of this will help clinicians to decide on treatment methodologies.
Finally, the developed Machine Learning Model (scoring system + classification method + type predictor) shall be deployed in hospitals at the discretion of doctors helping them better investigate patients for the presence of ovarian cancer.
The project initially aims to develop a scoring system for accurate prediction of ovarian cancer. Current scoring systems do not factor in a lot of recently observed attributes, resulting in high false positivity rate. Hence, to negate that, excessive testing needs to be done for which considerable amount of data is needed.
For developing the scoring system, the dataset must be prepared and balanced first. Though many data preprocessing techniques exist, they are not suitable for medical data because of the high sensitivity. Therefore, as part of our research work, we also aim to come up with data preprocessing techniques, specific to medical data.
Identifying what features affect presence of ovarian cancer in women is also a challenge which we aim to tackle by identifying the relevant features using various techniques.
The current prediction systems completely ignore borderline tumors though they are low-potential malignant tumors. Our research work also involves identifying borderline cases, as a third class, distinguishing them from benign and malignant tumors.
Project later aims to classify ovarian cancer into types like epithelial tumors, metastatic tumor and germ cell carcinoma tumors. Results of this will help clinicians to decide on treatment methodologies.
Finally, the developed Machine Learning Model (scoring system + classification method + type predictor) shall be deployed in hospitals at the discretion of doctors helping them better investigate patients for the presence of ovarian cancer.
Aims
1. To develop a technique to prepare medical data using preprocessing methods for further analysis.
2. To identify and select features relevant to predication of ovarian cancer using various feature selection methods.
3. To develop an efficient scoring system to predict Ovarian cancer.
4. To develop an efficient algorithm to classify the ovarian data set as benign, malignant and borderline.
5. To develop an algorithm to distinguish the malignant and borderline tumors as epithelial, germ cell, metastatic cancer or other types.
Collaborators
Dr. Geetha Maiya, Manipal Institute of Technology
Mr. Shyam Sundar Bharathi S, Manipal Institute of Technology
Mr. Ajay Rajendra Kumar, Manipal Institute of Technology