Image dataset from multiplex IHC stained TMA sections

IHC Cohort: Multiplex IHC stained histological slides were collected from Liaoning Cancer Hospital and Institute in China. The raw data collected are TMA sections, all of which are obtained from NSCLC patients. Non-overlapping image patches (256*256 pixels) are extracted from TMA sections and manually annotated by our collaborated pathologists via the Qupath software. Initially, seven TMA sections with multiplex stains including CD3, CD20, CD38, CDK4, Cyclin-D1, Ki67, and P53 were used. Except for CD3 stained TMA sections, we randomly cropped 25 image patches from each of these TMA sections, of which 17, 3 and 5 patches are correspondingly used for training, validation, and testing. Note that 18, 3 and 5 CD3 image patches are cropped for building dataset. To suppress over-fitting and enhance generalization of the SRSA-Net, we additionally included 81 annotated image patches from other five TMA sections with stains of CD34, CD68, D2-40, FAP, and SMA into training and validation set. In total, the IHC cohort includes 9,725 manually identified cell nuclei. The numbers of training, validation and testing image patches are 195, 36, and 35, respectively.

For the masks, the first channel and the second channel denotes the negative and positive nuclei pixels, respectively. Note that each pixel is labelled from 0 to n, where n is the number of individual nuclei detected. 0 pixels indicate background. Pixel values i indicate that the pixel belongs to the ith nucleus.  The last channel marks the information of all the nuclei pixels, where nuclei pixel are left as 0, otherwise 1.

fold1: training set

fold2: validation set

fold3: testing set