Keywords:
Metastases, Atelectasis, Screening, Computer Applications-Detection, diagnosis, Neural networks, Conventional radiography, CAD, Thorax, Lung, Cardiac
Authors:
N. Ramanauskas, J. Dementaviciene, J. Bialopetravičius, D. Barušauskas, J. Armaitis, J. Stankeviciene, G. Danys, R. Puronaite, R. Kizlaitis; Vilnius/LT
DOI:
10.1594/ecr2018/C-1896
Results
The obtained under receiver operating characteristic curve (AUC) testing scores for different pathologies (each versus the rest) are listed in table 1:
Table 1:
Label |
Freq |
AUC |
No Finding |
0.0815 |
0.8582 |
Cardiomegaly |
0.0027 |
0.7608 |
Effusion |
0.7385 |
0.8198 |
Atelectasis |
0.2683 |
0.6241 |
Pneumothorax |
0.0751 |
0.6925 |
Mass/Nodule |
0.0039 |
0.7266 |
Infiltration |
0.1330 |
0.6169 |
Pleural Thickening |
0.0045 |
0.6446 |
Emphysema |
0.029 |
0.7912 |
Fibrosis |
0.0661 |
0.7099 |
Edema |
0.0231 |
0.7695 |
Tuberculosis |
0.0574 |
0.4790 |
Average AUC with(out) TB |
0.7077 (0.7285) |
Highest AUC scores were obtained for labels no finding,
pleural effusion,
emphysema,
edema and cardiomegaly.
A possible explanation for this is that the definition of these labels is likely to vary the least in between the different datasets from different medical centers.
Furthermore,
as only frontal radiographs were used,
these are the pathologies that can be accurately identified using only frontal chest radiographs.
Whereas the pathologies that reached lower AUC scores (infiltration,
pneumothorax,
atelectasis,
tuberculosis) are more challenging to diagnose using only frontal chest radiographs and varies more in between different datasets.
We have developed a graphical user interface an (GUI) and an application programming interface (API) for visualization of the neural network output. The GUI / API displays a heat map which represents the localization of the detected label and the neural network confidence.
The API serves as a tool to identify the regions of an image by which the pathology was detected,
as well as pathologies.
Examples of the GUI (API output) are displayed in figures 1,
2,
3.
The API pipeline is briefly illustrated in a flowchart in figure 4. The accuracy of the API output closely correlates with the AUC scores for the detected labels.
Higher AUC labels are more likely to display an meaningful heat map.