Algorithm 1: Algorithm of difficult airway classification with DMF-Net
Input: multi-view data $\chi =\left\{X^{\left(1\right)},X^{\left(2\right)},X^{\left(3\right)},X^{\left(4\right)},X^{\left(5\right)}\right\};$ the label of data is $y=\left\{y_1,y_2,\dots ,y_N\right\}$.
Initialize: randomly initialize model parameters ${\theta }_{FEM},{\theta }_{MFFM},{\theta }_{HCAM}, {\theta }_{CPP},{\theta }_{CSP},{\theta }_{CLS}$.
While not converged, perform the following:
	1:Concatenate three front views as $x^1$ and two side views as $x^2$ on the channel dimension; 2:Obtain feature vectors $\{f_1^{gap},f_2^{gap}\}$ by utilizing the Feature Extractor Module, Multi-Scale Feature Fusion Module and Hybrid Co-Attention Module with Equations (1)–(4); 3:Obtain feature vectors ${\{z}_1^{cpp},z_2^{cpp}\}$$\mathrm{and} {\{z}_1^{csp},z_2^{csp}\}$ by utilizing the Complementarity Projection Head and Consistency Projection Head with Equations (5) and (6); 4:Concatenate $\{f_1^{gap},f_2^{gap}\}$ and input into the Classier Module to obtain final outputs $p$ with Equation (7); 5:Calculate $L_{focal}$ with Equations (9)–(11) and $L_{comple}, L_{consist}$ with Equations (12)–(14); then, calculate $L$ with Equation (8); 6:Update the parameters ${\theta }_{FEM},{\theta }_{MFFM},{\theta }_{HCAM}, {\theta }_{CPP},{\theta }_{CSP},{\theta }_{CLS}$ of the model by utilizing the Gradient Descent Algorithm;
end
Output: trained parameters ${\theta }_{FEM},{\theta }_{MFFM},{\theta }_{HCAM}, {\theta }_{CPP},{\theta }_{CSP},{\theta }_{CLS}$.