Home

Abstract

Colorectal carcinoma represents 15% of all cancers worldwide and its frequency is increasing in the EU. In Romania it ranks second as concerns cancer mortality. The early diagnosis of colon cancer significantly increases the chances to cure it. The aim of current project is to develop a medical information system accessible via Internet, dedicated for the diagnosis, grading and monitoring of patients with colorectal cancer. The program will comprise several modules and will be used to gather information of any type (numerical, text, images) about the patients from sections like clinical medicine, colonoscopy or anatomopathology and to store it on a server, so that it can be immediately available to all interested specialists. The data will consist of clinical and paraclinical indicators, observations from the colonoscopy exams, histopathology images and immunohistochemistry. Besides this practical tool that will ease the communication between different physicians, the software will comprise an intelligent component that will extract from the data set valuable information like correlations between different factors, weights that correspond to the importance of the indicators and disease diagnosis and grading. In traditional cancer diagnosis, the tumor analysis is made by the pathologist during the histopathology process. The histological study has several objectives like diagnosing the malignancy, identifying the histopathology variants, establishing the extent of the disease, determining the metastasis probability or the response to treatment. However, the analysis is subjective and it very much depends on the expertise and experience of the pathologist. To avoid inconsistency of the results and obtain objective judgments, through the current study, intelligent automated data mining tools that use precise quantitative measures will be developed and integrated in the envisaged software product that will give professionals the means to check and evaluate their reasoning.

Histopathological image analysis represents a very challenging and insufficiently explored research area and for developing successful diagnosis and grading tools within it, a close collaboration between pathologists, colonoscopy surgeons, medical clinicians and computer scientists is necessary. The software product helps in enhancing communication through the immediate synchronization of the data for all users and by constructing an accurate database through constant validation of the information received from the physicians. For computationally assessing a diagnosis for a histological image, the latter needs to be preprocessed, segmented, subject to the extraction of internal various attributes, while keeping only the valuable features to subsequently apply machine learning classifiers. There are many stages involved and for each phase there are numerous approaches that can be applied. We eventually aim to find the optimal settings for increasing the prediction accuracy in colorectal neoplasm diagnosis and grading.

As there is a worldwide acute need for benchmark datasets regarding histological images for colorectal diagnosis, we target the construction of such a data set that will be made available to scientists for research advancement in this domain.

The developed software product will be also made available to be used in hospitals: on the one hand, for patient monitoring, data gathering, enhancing communication between specialists from different sections, but also for computer assisted diagnosis. On the other hand, it will support histological image processing, as the dedicated tool will contain various possibilities for processing such images to be used in research purposes.