Code, dataset, and a detailed version of the paper "Chromosome Detection in Metaphase Cell Images Using Morphological Priors". (Accepted by IEEE JBHI)

A tensorflow (>2.0) project for chromosome detection in metaphase cell images

This is a one-stage detector for chromosome detection using skelecton-guided rotated anchors。

my enviroment

• Winows 10
• Anaconda python 3.7.3
• Tensorflow 2.8.0 with gpu
• cuda 11.6
• pytorch 1.12.0.dev20220504+cu116 (required for building the rotation libs, see path: tf_deep_karyotype/utils/rotation)

dataset

Data available at the baidu cloud:https://pan.baidu.com/s/1jxAbkKYKtGg-WKcceR9w0Q download code(提取码)：swcf

building rotation libs

[see the txt file at: tf_deep_karyotype/utils/how to build rotated_nms.txt ]

demo

(1)download checkpoint file from https://pan.baidu.com/s/1BWq8TP6y7ppqlHh4tqgFhQ      (download code: zm38)
(2)put the whole checkpoints dirctor to the tf_deep_karyotype
(3)open a cmd
(4)cd tf_deep_karyotype
(5) python demo.py

training

(1)download dataset from https://pan.baidu.com/s/1jxAbkKYKtGg-WKcceR9w0Q      (download code: swcf)
(2)put the dataset to your directory. 
(3)change the source path: data_root = r'D:\data\chromosome\labelme_cis_2022' to your dataset path('tf_deep_karyotype/scripts/labelme2mytraindata_converter.py')
(4)change the save path: save_root = r'D:\data\chromosome\chromosome_rotdet_v4 to your path ('tf_deep_karyotype/scripts/labelme2mytraindata_converter.py')
(5)run tf_deep_karyotype/scripts/labelme2mytraindata_converter.py to prepare data for training 
(6)run tf_deep_karyotype/run_train.bat to train the model