Cascaded-Recalibrated Multiple Instance Deep Model for Pathologic-Level Lung Cancer Prediction in CT Images
Algorithm 1
Cascaded-recalibrated multiple instance learning based on multiattribute features transfer for pathologic level lung cancer prediction in CT images.
Require:
(1)
The discovery group: , and the diagnosis group: ;
(2)
The set of 9 semantic attributes: ,, and the pathologic level lung cancer label: ;
(3)
For each attribute , all nodules on and were annotated with attribute score ; The pmal of each patient in was labeled with ,.
Ensure:
(4)
(1) The results of patient-level lung cancer prediction;
(5)
(2) Two-level cascaded recalibration coefficients, i.e., the contribution of the attributes to the nodule and the contribution of the nodules to the bag.//Step I: Attribute-specific modeling for each attribute on the discovery group .
(6)
for each in do
(7)
Jointly train feature extraction module and regression module with Equation (1).
(8)
end for//Step II: Recalibrated MIL modeling from each attribute to pmal on diagnosis group .
Recalibrate nodule-level features with equations (5) and (6)
(12)
Predict patient-level lung cancer with equation (7)
(13)
end for
(14)
Rank the 9 semantic attributes according to their independent performances on pmal.//Step III: Cascaded-recalibrated MIL modeling from top- attributes to pmal on diagnosis group .