Disease diagnosis system for supporting dual class, and method therefor

A method implemented in a system comprising a processor and a storage device and performed by a disease diagnosis system using a biometric image and a neural network having at least an initial layer, a final layer, and a layer preceding the final layer that is different from the initial layer, the method comprising steps of:

• receiving, by the neural network stored in the storage device, a given loss function and learning data labeled so that a given unitary unit included in the biometric image that includes a plurality of unitary units is to have at least one of a plurality of states;
• training to output a determination value corresponding to a probability having at least one of the plurality of states using the received learning data with respect to each unitary unit,
• the neural network comprises the initial layer, the final layer, and the layer preceding the final layer to output, by the final layer, a plurality of feature values corresponding to a probability that the unitary unit is to be determined as each of the plurality of states, and
• the loss function is defined to calculate a loss by incorporating both a first feature value corresponding to a first state and a second feature value corresponding to a second state, among feature values of the the initial layer, the final layer, and the layer preceding the final layer, into a dual labeling unitary unit labeled to have the plurality of states, comprising the first state determined to have a state of a higher probability and a second state determined to be lower than the first state, among multiple unitary units included in the biometric image,
• wherein the loss function is defined so that the second feature value has more losses than the first feature value,
• wherein the loss function is defined as Equation 1 below,
• L
• ⁡
• (
• x
• )
• =
• -
• log
• ⁢
• e
• x
• primary
• +
• p
• ·
• e
• x
• secondary
• ∑
• e
• x
• i
• ,
• [
• Equation
• ⁢
• 1
• ]
• wherein i indices the plurality of states, χi is a feature value corresponding to an i-th state among the plurality of states, χprimary is the first feature value, and χsecondary is the second feature value.