MSDA Capstone Final

Capstone Project for the Master of Science in Data Analytics

The aim of this project is the development of a computer vision algorithm that can correctly classify a given histology image into one of 27 given classes. Focuses of this project include: Using both Tile Level and Slide Level images:

1. Comparison of de novo neural network architectures, naïve Google Inception v3 architecture, and pre-trained Google Inception v3; the latter will involve reimplementation of the output layer of the network Using only Slide Level images:
2. Accurate Classification of DX and PM sections for a given tumor type
3. DifferentiationbetweenDXandPMsectionsfortumortypesotherthantheonesusedfortraining (generalizability) Using both Slide Level and Tile Level images:
4. Differentiation between tumor types that are known to look similar to humans. For example: a. Lung Adenocarcinoma vs. Lung Squamous Cell Carcinoma b. Kidney renal clear cell carcinoma vs. Kidney renal papillary cell carcinoma c. Glioblastoma multiforme vs. Brain Lower Grade Glioma