Deep-Learning-with-PyTorch

432 CHAPTER 14 End-to-end nodule analysis, and where to go next
14.6 What we see when we diagnose
Following along with steps 3a, 3b, and 3c in figure 14.16, we now need to run the full
pipeline from the step 3a segmentation on the left to the step 3c malignancy model on
the right. The good news is that almost all of our code is in place already! We just need
to stitch it together: the moment has come to actually write and run our end-to-end
diagnosis script.
1. Nodule Candidate Generation
1a. Segmentation
n 1b. Grouping 1c. sample tuples
((...,(..., IRC),
...
(..., IRC))
2. Nodule and malignancy claSsification
2a. Nodule claSsification 2b. ROC/AUC Metrics 2c. fine-tuning malignancy model
3. End-to-end detection
3a. (..., IRC) 3b. Is nodule? 3c. is Malignant?
Figure 14.16
The end-to-end project we are implementing in this chapter, with a focus on end-to-end detection
We saw our first hints at handling the malignancy model back in the code in section
14.3.3. If we pass an argument --malignancy-path to the nodule_analysis call, it
runs the malignancy model found at this path and outputs the information. This
works for both a single scan and the --run-validation variant.
Be warned that the script will probably take a while to finish; even just the 89 CTs
in the validation set took about 25 minutes. 7
7
Most of the delay is from SciPy’s processing of the connected components. At the time of writing, we are not
aware of an accelerated implementation.