Artificial intelligence for the detection of airway nodules in chest CT scans

Abstract

Objectives: Incidental airway tumors are rare and can easily be overlooked on chest CT, especially at an early stage. Therefore, we developed and assessed a deep learning-based artificial intelligence (AI) system for detecting and localizing airway nodules.

Materials and methods: At a single academic hospital, we retrospectively analyzed cancer diagnoses and radiology reports from patients who received a chest or chest-abdomen CT scan between 2004 and 2020 to find cases presenting as airway nodules. Primary cancers were verified through bronchoscopy with biopsy or cytologic testing. The malignancy status of other nodules was confirmed with bronchoscopy only or follow-up CT scans if such evidence was unavailable. An AI system was trained and evaluated with a ten-fold cross-validation procedure. The performance of the system was assessed with a free-response receiver operating characteristic curve.

Results: We identified 160 patients with airway nodules (median age of 64 years [IQR: 54-70], 58 women) and added a random sample of 160 patients without airway nodules (median age of 60 years [IQR: 48-69], 80 women). The sensitivity of the AI system was 75.1% (95% CI: 67.6-81.6%) for detecting all nodules with an average number of false positives per scan of 0.25 in negative patients and 0.56 in positive patients. At the same operating point, the sensitivity was 79.0% (95% CI: 70.4-86.6%) for the subset of tumors. A subgroup analysis showed that the system detected the majority of subtle tumors.

Conclusion: The AI system detects most airway nodules on chest CT with an acceptable false positive rate.

Key points: Question Incidental airway tumors are rare and are susceptible to being overlooked on chest CT. Findings An AI system can detect most benign and malignant airway nodules with an acceptable false positive rate, including nodules that have very subtle features. Clinical relevance An AI system shows potential for supporting radiologists in detecting airway tumors.

Keywords: Artificial intelligence; Lung neoplasms; Thorax; Tomography (X-ray computed); Tracheal neoplasms.

Fig. 1

Data selection flowchart. ^aPatients were pre-selected with ICD-O topology codes C339 and C340 before verification. ^bPatients were pre-selected with ICD-O topology codes C341–C349 and a natural language processing (NLP) analysis before verification (see Appendix 2). ^cPatients were pre-selected with an NLP analysis before verification (see Appendix 2). ^dNegative patients were sampled from 2004 to 2018 due to two-year follow-ups. ^ePoor-quality scans have severe breathing artifacts, missing slices, or partially depict airway nodules

Fig. 2

Examples of airway nodules in contrast-enhanced CT scans. For each panel, different window settings were applied for the lung (width: 1500 HU; level: −500 HU) and soft tissues (width: 350 HU; level: 0 HU). a A 60-year-old female patient with a carcinoid in the left lower bronchus (median attenuation value of 142 HU). b An 86-year-old male patient with a hamartoma with calcification in the left main bronchus (median attenuation value of 26 HU). c A 63-year-old male patient with nodular mucus plug in the distal trachea (median attenuation value of −18 HU). d A 57-year-old male patient with a lipoma in the left upper lobar bronchus (median attenuation value of −70 HU). HU, Hounsfield Units

Fig. 3

Examples of tracheobronchial tumors, ranked by subtleness. Obvious tumors: a A 73-year-old female patient with an adenosquamous carcinoma in the left lower lobe bronchus with evident post-obstruction atelectasis. b A 67-year-old male patient with a squamous cell carcinoma in the right lower lobe bronchus showing extensive extraluminal growth. Relatively obvious tumors: c A 67-year-old female patient with an adenoid cystic carcinoma in the trachea. d A 49-year-old male patient with an adenoid cystic carcinoma in the left main bronchus. Subtle tumors: e A 69-year-old male patient with a squamous cell carcinoma in the right main bronchus, not reported in the original examination. f A 62-year-old male patient with a squamous cell carcinoma in the trachea. Very subtle tumors: g A 65-year-old male patient with a squamous cell carcinoma at the ostium of the right lower lobe bronchus, not reported in the original examination. h A 60-year-old female patient with a hamartoma in the apical segment of the right upper lobe bronchus

Fig. 4

Free response receiver operating characteristic (FROC) curves of the AI system. The solid and dashed lines correspond with the observed sensitivities and mean sensitivities based on bootstrap samples, respectively. The shaded areas represent the 95% CIs. a FROC curves for all nodules and the subgroups tumors (benign and malignant) and malignant tumors only. b FROC curves for the tumor subcategories as presented in Table 3. The number of detected tumors is displayed near the FROC curves

Fig. 5

Examples of true positives of the AI system. Non-tumors: a A 72-year-old female patient with a mucus plug in the right main bronchus. b A 51-year-old male patient with a mucus plug in the trachea bifurcation, left the main bronchus. c A 64-year-old male patient with a mucus plug in the proximal trachea. d A 68-year-old male patient with a mucus plug in the right lower lobe bronchus. Tumors: e A 68-year-old male patient with a squamous cell carcinoma at the ostium of the right upper lobe bronchus. f A 63-year-old male patient with a squamous cell carcinoma at the ostium of the right lower lobe bronchus, not reported in the original examination. g A 42-year-old female patient with an adenoid cystic carcinoma of the trachea and right main bronchus. h A 50-year-old male patient with a carcinoid in the left lower bronchus

Fig. 6

Examples of false negatives and positives of the AI system. False negatives: a A 58-year-old female patient with a metastasis of a Hurthle cell carcinoma in the apical segment of the right upper lobe bronchus. b An 85-year-old male patient with a non-small cell carcinoma in a subsegmental bronchus in the left upper lobe. c A 68-year-old male patient with a squamous cell carcinoma in a subsegmental bronchus in the left upper lobe, not reported in the original examination. d A 67-year-old male patient with a duct carcinoma at the ostium of the left upper lobe bronchus. False positives: e A 78-year-old male patient with non-nodular mucus secretion in the trachea. f A 28-year-old male patient showing anterior bowing of the posterior trachea membrane. g A 23-year-old male patient with a trachea tube. h A 75-year-old male patient with detection on the left inferior pulmonary vessel