GI Tract Disease Classification System for Patients and Doctors
Gastrointestinal endoscopy has been an active field of research due to acute GI Tract diseases and misdiagnoses of these diseases. This project aims to predict Gastrointestinal Tract diseases using images and videos. The end-product of this project will be a website. The website will encompass two p
2025-06-28 16:27:29 - Adil Khan
GI Tract Disease Classification System for Patients and Doctors
Project Area of Specialization Computer ScienceProject SummaryGastrointestinal endoscopy has been an active field of research due to acute GI Tract diseases and misdiagnoses of these diseases. This project aims to predict Gastrointestinal Tract diseases using images and videos. The end-product of this project will be a website. The website will encompass two portals: a Patient's portal and a Doctor’s portal.
The Patient's Portal will enable the patients to upload their past reports, endoscopy images, and videos to maintain a record of their past tests, which will reduce the time it takes for another doctor to evaluate the patients as they will have all their previous records on hand available. Furthermore, it will also include an option for the patient to upload their prior reports in the system database. Hence, the patient record will be accessible from anywhere through this system.
The Doctor's Portal will allow doctors from different hospitals to upload images and videos from endoscopy for diagnosis. Furthermore, the system will also provide the option for analyzing the images. The doctor can also view the reports for its patients and can give feedback for those reports. The analysis of multimedia data obtained from endoscopy will utilize Machine Learning and Deep Learning models to predict the diagnoses of GI Tract diseases. Active Learning will also be employed to further improve the accuracy of the models through feedback provided by the specialized doctor.
The project will follow an iterative approach for development. The end-product of this research will aid doctors in diagnosing GI tract diseases by observing the irregularities in the images and preventing misdiagnoses to a decent extent. It will also lower the inherent miss rate in GI tract disease diagnosis.
This project aims to assist doctors with the quicker diagnosis of Gastrointestinal Tract diseases by solving multi-class disease identification problems in the gastrointestinal tract using multimedia data such as images and videos. The images are easily captured by using capsule endoscopy or the traditional endoscopy method, where a thin tube with a camera is inserted through the patient's mouth to get the screening.
Some principal objectives of this project include:
- To research multiple textural features for the detection of various GI Tract diseases and ensemble these features together to improve overall accuracy.
- To provide speedy results of the acquired multimedia data, as currently most of the doctors are not experienced enough to detect diseases.
- To develop a web-based application where hospitals from Pakistan can upload multimedia data acquired from traditional/capsule endoscopy to test our system.
- To acquire a user base of specialized doctors to approve the predicted result and further improve the accuracy of our designed solution.
- To develop a system where patients can upload their previous reports. Thus when they go to another hospital, it is feasible for the doctor to review their past tests and deduce results based on their experience.
Texture is a significant and defining feature for identifying anomalies in images, which help discriminate against one another. Textural features such as Local Binary Pattern [5], Local Ternary Pattern [7], and Haralick descriptors [9] can be employed to identify abnormality, along with histogram color descriptors. For deep learning, transfer learning approaches on pre-trained models such as ResNet, VGGNet, ImageNet, etc. will be explored with modified final dense layers, and the model which holds the best overall specificity and cross-validation accuracy will be selected for the final product [1]. Active learning will also be employed to improve the model's accuracy.
The proposed system will be utilized for concluding the final disease verdict. Different hand-crafted and deep features will be extracted, and ensembles of those features will be used for prediction. Afterward, a specialized doctor will validate whether the prediction is correct or not. If the prediction is incorrect, the program will use active learning to query the model to adjust weights for future predictions.
The proposed system will encompass two portals: a Patient's portal and a doctor’s portal.
The Patient's Portal will enable the patients to upload their past reports, endoscopy images, and videos to maintain a record of their past tests, which will reduce the time it takes for another doctor to evaluate the patients as they will have all their previous records on hand available. Furthermore, it will also include an option for the patient to upload their prior reports in the system database. Hence, the patient record will be accessible from anywhere through this system. Figure 2 shows the functionality in the Patient's portal.
The end product of this research will aid doctors in diagnosing GI tract diseases by observing the irregularities in the images and prevent misdiagnoses to a decent extent. It will also lower the inherent miss rate in GI tract disease diagnosis.
Capsule endoscopy data requires being sent to external parties for screening due to lack of expertise, which causes delays in results. It is possible to mitigate this problem if the diagnosis can be performed on-site in the hospital, which will reduce the overall result time.
About 20% of the whole population of Pakistan has access to safe drinking water. The remaining 80% of the population is compelled to use unsafe drinking water due to the scarcity of safe and healthy drinking water sources. The primary sources of contamination are sewerage and the dumping of leftover chemicals from industries. Due to these reasons, stomach disease is very widely prevalent among people in rural areas. In some rural areas, the equipment is available for GI tract diagnosis, but the doctors in these rural areas are not experienced enough to diagnose such diseases. Thus, patients in these areas prefer to travel to big cities like Karachi to get their diagnosis. As most patients come from rural areas, they cannot afford to travel back and forth. Therefore, this system will prevent the wastage of time of patients by diagnosing the problem earlier and avoid unnecessary tests because of the availability of previous tests in the database. Hence it will help towards lowering the patients' overall expenses.
This project comprises two phases. One, the Research Phase, while the other, the Development Phase. In the research phase, different textural features will be explored to identify those features which produce the most accurate results. In the development phase, the system will be developed as a web application containing two portals, the Patient's Portal and the Doctor's Portal.
Through the Patient's Portal, patients will be able to upload their previous and new records and provide access to a doctor to view them.
Through the Doctor's Portal, a doctor will be able to view a patient's previous record, upload endoscopy images and videos, and view and verify the prediction of the system regarding the GI Tract diseases.