A Comprehensive Survey on the Progress, Process, and Challenges of Lung Cancer Detection and Classification

Table 14

Challenges and research directions for lung nodule and cancer diagnosis.

Name	Challenges	Research direction
Insufficient number of annotated medical datasets with cases	All datasets are not publicly available	All datasets need to be available openly. Additionally, research should be conducted utilizing such imaging modalities using unpublished datasets. All datasets should be disclosed for future research works and implementations.
Accurate segmentation	Segmentation models are not properly executed	All segmentation models need to be implementing in various modalities which may uplift the lung nodule and cancer detection results
Nodule size and types	Small nodules are needed to be detected more efficiently	All kinds of nodules need to be investigated. Implementing feature extraction and selection can detect most of the nodules. Nodules can be identified by feature and classifier selection.
Efficient CADe system	Nodules and cancer detection need to be more accurate using all architectures	Random forest, SVM, DBN with RM, and CNNs are mostly used for lung cancer diagnosis. ML and DL networks of other kinds should be analyzed in this field.
Volumetric measurements	All lung image shapes are not the same. So, all datasets need to be extracted.	When patients are breathing, their lung shape changes and it varies from patient to patient. We recommend investigating all datasets and measuring different shapes of lungs.