Figures for Demo of 'A semi-supervised image segmentation framework for extracting breast tumors from Magnetic Resonance images' (CIKM2017 Demo Draft)

Figure 4: Dataset summary of Breast MRIs. Our dataset includes a set of 2D MRI records of 21 patients, which are in uint8 format. Figure 4 (on Website ) shows the tumor information in the dataset, where 14 patients have mass malignant tumors(M), and 7 have non-mass tumors (NM). Fourteen patients have infiltrating ductal carcinoma (IDC) tumor, two have the ductal carinoma in situ of the breast (DCIS), and five have both IDC and DCIS tumors. In the column of 'TumorLayers(TL)', we list the layers of MRIs containing tumors of each patient. 'TLNum' column shows the number of layers containing tumors. The total number of TL is 407 of the 21 patients

Figure 5: Screenshot of an MRI segmentation process of SSTS. Users of the system can change parameter settings to see different segmentation results. The DB-pieces are classified by a pre-trained classifier with respect to a parameter setting (e.g. t = 08, s = 10, d = 10, r = 0.5)

Figure 6: Examples of tumor extraction results of SSTS, FCM, and Multi-level thresholding (MT) on MRIs of two patients: MRI of patient 1 contains a non-mass tumor, and MRI of patient 2 contains an mass tumor

Figure 7: Qualitative evaluation of Mass tumor extraction of MRIs of nine patients based on SSTS, FCM, and MT. The parameters of SSTS are set as: <t,s,d,r> = <0.8,5,20,0.6> (see Table 2 in this paper)

Figure 8: Qualitative evaluation of Non-mass tumor extraction of MRIs of seven patients based on SSTS, FCM, and MT. The parameters of SSTS are set as: <t,s,d,r> = <0.7,10,8,0.6> (see Table 2 in this paper)