Differential diagnosis of bipolar II disorder and major depressive disorder: Integrating multimodal approaches to overcome clinical challenges

Table 3 Characteristics and application scenarios of different machine learning algorithms in bipolar II disorder vs major depressive disorder differential diagnosis

Algorithm type	Advantages	Limitations	Application scenarios	Example applications	Ref.
Support Vector Machine	Performs well on small, high-dimensional datasets; clear decision boundaries	Sensitive to parameter tuning; high computational cost for large datasets	Gene expression analysis; language pattern recognition	Distinguishing emotional text features between BD-II and MDD	[51,54,55]
Random Forest	Strong robustness; handles nonlinear relationships; provides feature importance ranking	May overfit small datasets	Multi-modal feature integration; questionnaire + imaging data	Gene combination screening; questionnaire-based classification	[52,53]
Deep Learning (CNN, RNN)	Automatically extracts complex features; suitable for image and text data	Requires large datasets and high computational power; low interpretability	Facial expression analysis; social media data	Micro-expression recognition; emotion classification	[9,52,55,56]
Linear Discriminant Analysis	Simple and efficient; high interpretability	Limited by linear assumptions; unsuitable for highly nonlinear data	Emotional vocabulary frequency analysis	Vocabulary-based emotion classification	[54,55]
Polygenic Risk Score	Integrates genetic information; enables personalized risk prediction	Depends on large-scale genomic data	Early risk prediction; pediatric cohort studies	Predicting future risk of BD development	[10]
Hybrid Models	Combines strengths of multiple algorithms; improved performance	Complex implementation; requires coordination between models	High-dimensional, multi-modal data analysis	SVM + RF combined diagnosis	[55,57,58]

MDD: Major depressive disorder; SVM: Support vector machines; BD-II: Bipolar II disorder; BD: Bipolar disorder; RF: Random forest.