Table 1.

Results of the DNN models for combinations of MRI sequences.

Combination set	SS-Avg (%)	Accuracy (%)	Sensitivity (%)	Specificity (%)	ADC	DWI	dynamicaxi	dynamicsag	fsT1axi	fsT1CEaxi	fsT1CEcor	fsT1CEsag	fsT1sag	fsT2axi	T1axi	T1sag	T2axi	T2cor	T2sag
Combination set1	91.3	89.9	88.7	94.0		●											●		●
Combination set2	91.3	90.5	89.8	92.9			●	●	●	●	●	●	●	●	●	●	●	●	●
Combination set3	91.1	91.5	91.9	90.3	●	●	●	●	●		●	●	●	●		●	●	●	●
Combination set4	91.0	89.3	87.8	94.2	●									●					●
Combination set5	90.8	90.7	90.7	90.9		●	●	●	●	●	●		●	●		●	●	●
Combination set6	90.5	90.5	90.5	90.5		●	●	●	●	●	●		●	●		●	●	●	●
Combination set7	90.5	89.4	88.5	92.5	●	●	●	●	●	●	●	●		●		●	●	●	●
Combination set8	90.5	90.0	89.6	91.3	●		●	●	●	●	●	●	●	●		●	●	●
Combination set9	90.4	89.7	89.0	91.8		●	●	●		●	●	●	●	●	●	●	●	●	●
Combination set10	90.3	91.1	91.9	88.8		●	●	●	●	●	●	●	●	●	●		●	●	●
Average	90.8	90.3	89.8	91.7

The top 10 combinations of MRI sequences in terms of the SS-Avg (average of sensitivity and specificity) are shown. The average data for the top 10 combination sets were adopted as the results of the DNN models (accuracy: 90.3%; SS-Avg: 90.8%; sensitivity: 89.8%; specificity: 91.7%). Combination set 1 (T2axi, T2sag, and DWI) was the most accurate combination. DWI: diffusion-weighted imaging.