IHC Cell Counts with Crowd Counting CNN

Objective: Utilize a crowd counting CNN model to perform fluorescent cell counts that bounding-box-based NNs cannot quantify.

    ❖ For the general (cytoplasmic & nucleic) model, MAE = 79.8 (n = 58)
        ❖ for cell counts of ~400.
    ❖ NOTE: Lower cell-count images fared better: 
        ❖ ie. an image with 62 cells had a prediction of 60 cells. 

SKILLS OBTAINED

General Skills:

    ❖ Convolutional Neural Networks
        ❖ Rectified Linear Unit
        ❖ Batch Normalization
        ❖ 2D CNN
    ❖ CUDA processing
    ❖ Hierarchical Data Format (.h5)
    ❖ Data Annotation w/ MatLab

Languages:

    ❖ Python 3
    ❖ JSON
    ❖ MATLAB

Frameworks:

    ❖ Pytorch (CUDA)

Other Skills:

    ❖ Git/Version Control

Project Sample