Team: Department of Electronics and Communication engineering, NSSCE Palakkad Kerala, India
Authors: Prof. Nandakumar P, Sarath PC, Vishnu Prasad M, Yadukrishnan M, and Sreejith Valsan M
In our work we have followed a heuristic approach of finding tumorous tiles from non-tumorous tiles and thereby deciding whether the given tile is tumorous or not. Here we have used several features such as mean intensity value, different counting algorithms (object counting, shape counting and cell counting) and some features like peak-to-rms ratio, peak to peak ratio etc., of featured curves and histogram of the image tile. These features were extracted for a set of training (both Tumor and Normal WSIs) and studied. It was observed that these features followed a certain fashion for each kind of tiles i.e., tumorous, non-tumorous, blank, edge etc. The unwanted classes of tiles could be removed by appropriately setting the value for one or more of the features. Due to system memory limitations, instead of processing the WSI as whole we took advantage of the feature of tiff files and processed the image tile by tile.
The algorithm uses several image preprocessing techniques like grayscale conversion, median and Gaussian filtering etc. The algorithm employs a number of screening for each tile using the above mentioned features. The algorithm is maximum tapered to the end where it removes all of the unwanted class of tiles that may get in despite all the screening done in the earlier stages. For this we have used a number which is a combination of the some of the above mentioned features.
The following figure shows the receiver operating characteristic (ROC) curve of the method.
The following figure shows the free-response receiver operating characteristic (FROC) curve of the method.
The table below presents the average sensitivity of the developed system at 6 predefined false positive rates: 1/4, 1/2, 1, 2, 4, and 8 FPs per whole slide image.
Error rendering graph from file