Journal of Cancer Diagnosis
Open Access

Lung cancer; Deep learning; Tumor detection; Diagnosis; Healthcare; Medical technology

Introduction

The number of lung cancer patients globally has increased, and the incidence rate has gone up every year, due to the deterioration of the environment brought on by increasingly severe air pollution and variables like smoking and occupational exposure. Every year, over 60% of the 1.4 million lung cancer cases reported worldwide will undergo examination. Death within the first year of life; the desired five-year survival rate is 15%. Today, lung cancer is among the cancers with the greatest incidence and fatality rates worldwide. Concerns have been raised about how to correctly detect lung cancer and stop its occurrence and progression [1].

Godfrey Hounsfield, a British electronic engineer, created the first computed tomography (CT) gadget for brain imaging at the beginning of the 1970s, and since then, CT technology has been widely used. CT, the most popular imaging method in medicine, is frequently used to find lung lesions. High density resolution in CT images allows for the creation of contrast pictures even for parts with minute density variations, such as soft tissues in humans. But as imaging technology advances and clinical needs rise, particularly with the introduction of high-resolution CT technology, the volume of medical imaging data is expanding quickly [3]. According to statistics, 90% of the information stored in hospitals is in the form of imaging data, and as medical technology advances, this percentage is growing 30% year. The number of qualified imaging diagnostic doctors has only grown by 4% at the same period. A whole lung CT scan session typically consists of 150 to 300 pictures. Radiologists now do difficult diagnostic procedures, which tests both their physical and mental stamina. It demonstrates how reading time for a single scan sequence increases while reading accuracy decrease [4]. After staring at CT pictures for a while, human eyes are very likely to grow tired, which can result in missed diagnoses and incorrect diagnoses. The research team at Johns Hopkins University in the United States conducted related studies and discovered that there is a 30 percent chance that a single imaging specialist will overlook the shadow of clinically relevant lung nodules while diagnosing a chest CT. Therefore, it is imperative that doctors use computers to aid in reading and diagnosis, enhancing the speed and precision of the latter [5] (Table 1).

Table 1: This table shows a selection of studies that have applied deep learning models to diagnose lung cancer through medical imaging.

Computer-aided diagnostic (CAD) technology has made significant strides recently thanks to the quick growth of computer hardware, software, and deep learning technologies. It has also gradually proven its clinical worth in diagnosis [5]. Clinical medical diagnosis specialists are starting to use CAD software more frequently to aid in diagnosis. The hospital’s workflow eventually incorporates CAD software as a “second reader”. Doctors’ film reading diagnoses are now much more quickly and accurately thanks to medical image processing technologies and computer analysis capabilities [7].

The artificial neural network is a mathematical representation of the human brain’s neuron network created by an abstract simulation from the viewpoint of information processing, transmitting, and receiving. In the fields of engineering and academia, it is frequently referred to as a neural network. Deep learning technology, an improved version of neural networks, has gradually solved many practical problems in the fields of pattern recognition, speech processing, biomedicine, economic forecasting, etc., and has demonstrated exceptional performance in recent years with the continuous deepening of related research work [8]. Innovative Features the field of medical auxiliary diagnostics has gradually started to utilise this technology as a result of the widespread use of technologies like big data and artificial intelligence in the medical industry. Deep learning technology has advantages in picture segmentation, classification, and detection when compared to more conventional approaches like probability and statistics [9] (Table 2).

Table 2: This table shows a selection of applications of deep learning technology across different industries.

Higher-Level Performance In addition, areas of medical image processing and analysis like picture segmentation and registration make extensive use of neural network technology. As part of deep learning technologies, recurrent neural networks (RNN) and long short-term memories (LSTM) are also employed in speech and text processing for medical diagnosis [10]. Deep learning technology has many uses, is effective, and is simple to understand; therefore one of its most common applications is in the field of medical care [9]. As a result, in the context of the study’s potential medical applications, we created an intelligent medical detection algorithm to help physicians identify well-known organs and disorders, considerably increasing the accuracy of their diagnosis. In this study, we tested the efficacy of a deep learning-based model using CT scans of lung cancer to accurately and promptly detect long illness [11].

Discussion

The use of computer-aided diagnosis in the field of medical imaging has developed significantly over the past several years, and new CAD technologies have appeared one after the other. Particularly, the development of CAD systems based on deep learning techniques has contributed to a decline in the percentage of missed diagnoses among clinicians and an increase in diagnosis accuracy. New techniques for lung cancer imaging detection are also constantly being developed at the same time [12].

In 1985, the fundamental idea of CAD was first put forth. There has been work done on CAD research and development. Two components make up the CAD application for lung cancer: computer-aided detection and computer-aided diagnosis. The primary purpose of the former is to help radiologists identify and find lung cancer symptoms, whereas the primary purpose of the latter is to help radiologists identify probable lung cancer tissues that have already been identified [13]. The Maximum Likelihood Model, Bayes’ Theorem, and other early CAD mathematical models that focus on diagnosis of benign and malignancy are the most well-known. This kind of CAD system merely employs computers to carry out some basic processing of medical data as an expert system based on rules and knowledge. The features throughout the entire process have been designed by humans, thus there will be flaws in the test findings [14]. The local shape and curve features of the image are used to discover potential structures in the lung data volume, followed by two sequential K-nearest neighbour classifiers to reduce false positives. Other CAD systems for lung nodule detection, such ETROCAD and Large CAD, are still using outdated techniques. Later, efforts were undertaken to create an intelligent CAD system using artificial neural networks’ potent learning capabilities, and some progress was made in this direction [15].

Intelligent CAD solutions significantly outperform conventional CAD systems by eliminating their reliance on artificially created features and realising autonomous knowledge acquisition and adaptive feature reasoning. For instance, M5LCAD and MOT_M5Lv1 both implement some system tasks using an artificial neural network, which significantly increases the nodule detection rate and accuracy. The Ali Health team’s “Doctor You” CT lung nodule intelligence detection technology successfully read approximately 9,000 images in 30 minutes for home use, with a recognition accuracy of more than 90% [16]. The “Tencent Miying” technology, which Tencent Youtu Lab formally unveiled in August 2017, has an accuracy rate of 95% for detecting lung nodules and an 80% recognition rate for early-stage lung cancer. It is capable of finding nodules as small as 3 mm and larger. The ranking list of “Lung Nodule Detection” and “False Positive Reduction” in the reputable assessment LUNA16 (LungNoduleAnalysis2016) in the area of global medical imaging has been updated successively by Jianpei CAD released. The world record set by the mission. Currently, pulmonary nodular lesions can be detected and diagnosed effectively using CAD technology. Some international research findings have even been used in clinical practise after being verified by the US Food and Drug Administration [17].

The demand for data is intensifying due to the present neural network’s quick development. The Lung Image Data Base Consortium Image Collection (LIDC-IDRI) and LUNA16 are two of the most popular open-source lung CT data sets at the moment. The American National Cancer Institute created the lung slice data set known as LIDC-IDRI. Chest medical imaging files and the related diagnosis result annotation files make up the majority of it. This data set’s goal is to investigate the early stage of lung nodules as shown by their characteristics [18]. Characteristics of cancer 1018 study examples in all are included in this data set. Four thoracic radiologists with extensive experience interpreting pictures provided a two-stage diagnostic and annotation for the images in each scenario [19]. The labelling information also includes the nodule’s properties, such as its sphericity, calcification, benignity, and malignancy, in addition to its contour. The evaluation of the nodule is aided by these characteristics. The Grand Challenges platform’s Lung Nodule Detection competition, LUNA16, also includes an open-source lung CT data set. Based on the LIDCIDRI lung nodule data set, 888 lung CT scans were created by screening the data set for slice thickness, spatial continuity, completeness, marked nodule size, and the number of marked physicians. It is a dataset with images and annotations for nodules [20].

Conclusion

The use of deep learning technology on CT images of thyroid and lung cancer was thoroughly examined in this paper. It primarily achieves the detection of thyroid and lung nodules, False Positive Reduction, benign and malignant classification, benign and malignant classification of the lung, and False Positive Reduction. A network named Nodule-Net, which combines the U-Net and RPN networks, has been proposed for the task of identifying lung nodules. Furthermore, the test set performance of the suggested model was close to perfect. It is also quite competitive when compared to other LUNA 16 competition solutions; the test set’s final average FROC value is 0.876. Whether it is used for data labelling or preliminary screening, the proposed model is the best option.

It just takes 0.5 seconds to discover a nodule because to the extremely quick detection speed. Weighted cross-entropy is utilised as a loss function in the 3D U-Net system to balance the imbalance of positive and negative data in the False Positive Reduction task for lung nodules. The final test set’s FROC value is 0.883, which is likewise a significant advantage over previous LUNA16 solutions. In the process of classifying lung nodules as benign or malignant, a 3D U-Net structure is utilised as the network structure for extracting features, and four fully linked structures—Feature Combine, MaxP, Noise-or, and Leaky Noise-or—are employed to calculate the final benign and malignant classifications. The Leaky Noise-or structure among them offers the best classification precision. The classification network’s antiinterference performance is somewhat enhanced by this topology.

Conflicts of Interest

None

Acknowledgement

None

1. Santos IP, Barroso EM, Schut TCB (2017) Raman spectroscopy for cancer detection and cancer surgery guidance: translation to the clinics. Analyst 142: 3025-3047.

Google Scholar, Crossref, Indexed at

2. Jermyn M, Desroches J, Aubertin K (2016) A review of Raman spectroscopy advances with an emphasis on clinical translation challenges in oncology. Physics in Medicine and Biology 61: 370-400.

Google Scholar, Crossref, Indexed at

3. Yue S, Cheng JX (2016) Deciphering single cell metabolism by coherent Raman scattering microscopy. Current Opinion in Chemical Biology 33: 46-57.

Google Scholar, Crossref, Indexed at

4. Sun J, Zhang J, Lu J, Gao J, Ren X, et al. (2016) BRAF V600E and TERT Promoter Mutations in Papillary Thyroid Carcinoma in Chinese Patients. PLoS One 11: 153-319.

Google Scholar, Crossref, Indexed at

5. Jin L, Sebo TJ, Nakamura N, Qian X, Oliveira A, et al. (2006) BRAF Mutation Analysis in Fine Needle Aspiration (FNA) Cytology of the Thyroid. Diagn Mol Pathol 15.

Google Scholar   Crossref, Indexed at

6. Ferrari M (2005) Cancer nanotechnology: opportunities and challenges. Nat Rev Cancer 5: 161-171.

Google Scholar , Crossref, Indexed at

7. Miller AD (2003) the problem with cationic liposome/micelle-based non-viral vector systems for gene therapy. Curr Med Chem 10: 1195-1211.

Google Scholar, Crossref, Indexed at

8. Cho K, Wang X, Nie S, Chen ZG, Shin DM, et al. (2008) Therapeutic nanoparticles for drug delivery in cancer. Clin Cancer Res 14: 1310-1316.

Google Scholar , CrossRef, Indexed at

9. Davis ME, Chen ZG, Shin DM (2008) Nanoparticle therapeutics: an emerging treatment modality for cancer. Nat Rev Drug Discov 7: 771-782.

Google Scholar, Crossref, Indexed at

10. Lammers T, Hennink WE, Storm G (2003) Tumour-targeted nanomedicines: principles and practice. Br J Cancer 99: 392-397.

Google Scholar, CrossRef, Indexed at

11. Haka AS, Volynskaya Z, Gardecki JA (2006) In vivo margin assessment during partial mastectomy breast surgery using Raman spectroscopy. Cancer Res 66: 3317-3322.

Google Scholar, Crossref, Indexed at

12. Pectasides D, Pectasides M, Nikolaou M (2005) Adjuvant and neoadjuvant chemotherapy in muscle invasive bladder cancer. Eur Urol 48: 60-67.

Google Scholar, Crossref, Indexed at

13. Mitchell E, Macdonald S, Campbell NC, Weller D, Macleod U, et al. (2008) Influences on pre-hospital delay in the diagnosis of colorectal cancer. Br J Cancer 98: 68-70.

Google Scholar, Crossref, Indexed at

14. Singh H, Daci K, Petersen LA (2009) Missed opportunities to initiate endoscopic evaluation for colorectal cancer diagnosis. Am J Gastroenterol. 104: 2543-2554.

Google Scholar, Crossref, Indexed at

15. Ferrari M (2005) Cancer nanotechnology: opportunities and challenges. Nat Rev Cancer 5: 161-171.

Google Scholar , Crossref, Indexed at

16. Miller AD (2003) the problem with cationic liposome/micelle-based non-viral vector systems for gene therapy. Curr Med Chem 10: 1195-1211.

Google Scholar, Crossref, Indexed at

17. Davis ME, Chen ZG, Shin DM (2008) Nanoparticle therapeutics: an emerging treatment modality for cancer. Nat Rev Drug Discov 7: 771-782.

Google Scholar, Crossref, Indexed at

18. Lammers T, Hennink WE, Storm G (2003) Tumour-targeted nanomedicines: principles and practice. Br J Cancer 99: 392-397.

Google Scholar, Crossref, Indexed at

19. Bodanese B, Silveira FL, Zangaro RA, Pacheco MTT, Pasqualucci CA, et al. (2012) Discrimination of basal cell carcinoma and melanoma from normal skin biopsies in vitro through Raman spectroscopy and principal component analysis. Photomedicine and Laser Surgery 30: 381-387.

Google Scholar, Crossref, Indexed at

20. Frank CJ, Redd DCB, Gansler TS, McCreery RL (1994) Characterization of human breast biopsy specimens with near-IR Raman-spectroscopy. Analytical Chemistry 66: 319-326.

Google Scholar  , Crossref, Indexed at