Lung disorders remain at the centre of the burden affecting health on a global scale, and long-term respiratory complications caused by COVID-19 or other pulmonary disorders compound the burden. The accurate, timely identification of these disorders can improve the treatment outcome and diminish the number of fatalities. In recent years, the combination of deep learning techniques and feature selection methods has shown great promise in improving the accuracy and efficiency of medical imaging systems in diagnosis. This paper reports a systematic comparative review of the recent deep learning approaches to lung disorder diagnosis. The numerous works based on model architecture, feature selection strategies, datasets, performance metrics and clinical limitations are compared. In contrast to the current surveys that are largely focused on only a single disorder or a specific architecture, this review compares architectures, hybrid architectures and feature selection frameworks for the various lung disorders in a unified manner. Different models, such as DenseNet-121, Xception, InceptionV3, CNNs, Vision Transformers, MobileNet, RVCNet, EfficientNetB0, CNN-LSTM, XGBoost, Random Forest, and Decision Trees, are investigated. The results show that lightweight architectures like MobileNet and RVCNet can generally achieve better efficiency and accuracy, while more complex architectures and hybrid CNN-LSTM architectures can enhance the predictive performance. However, there are several challenges that are not yet addressed, such as mislabelling of public datasets, overfitting, lack of generalisation to unseen clinical data, lack of interpretability of complex architectures, and lack of ability to handle patients with pre-existing lung abnormalities. Also, multi-label classification systems are yet weak for overlapping pulmonary disorders. These challenges underline the need for trusted datasets, easily interpretable hybrid frameworks and clinically usable, intelligent diagnostic systems for future works in lung disorder detection.