Page 105 - TD-3-3

Tumor Discovery Effectiveness of AI imaging for lung nodules

group, the AUC was 0.737, with a sensitivity of 83.19% showed higher specificity than the AI group (Figure 1
(95% CI: 0.7766 – 0.8782) and a specificity of 66.15% and Table 5).
(95% CI: 0.5335 – 0.7743). Both the AI group and the
physician-reading group demonstrated good predictive 4. Discussion
value for the qualitative diagnosis of lung nodules. The Primary bronchopulmonary cancer, commonly known
AI group exhibited higher sensitivity compared to the as lung cancer, is a globally significant malignant tumor
physician-reading group, while the physician group with high mortality rates. In 2022, lung cancer was
28
the most common malignancy in China, accounting
Table 2. Post‑operative pathological test results of for 18.06% of new cases and 23.9% of fatalities.
29
pulmonary nodules With projections indicating a 50% increase in cancer
burden by 2040 compared to 2020, early screening and
Lesion type Benign group Malignant
(quantities) group diagnosis are becoming increasingly crucial. Advances
(quantities) in precision medicine, such as targeted therapies and
Inflammatory pseudotumor 39 - immunotherapies, have markedly improved outcomes
30
Hamartoma 9 - and survival rates for lung cancer patients. Nonetheless,
Organized pneumonia 8 - early detection remains essential, as the prognosis of
lung cancer heavily depends on the stage of the disease.
Tuberculosis 5 - For example, the 5-year survival rate for stage IA lung
Bronchogenic cyst 1 - cancer can reach 85%, while it drops to only 6% for
Sclerosing pneumocytoma 1 - stage IV. Early-stage lung cancer often lacks obvious
31
Mycotic infection 1 - symptoms, leading to most cases being diagnosed at
Benign lesion 1 - advanced stages, where the 5-year survival rate falls
Adenocarcinoma - 215 below 20%. Currently, imaging plays a critical role in
32
Squamous carcinoma - 5 the early screening and diagnosis of lung cancer, with
Poorly-differentiated carcinoma - 3 early-stage lung cancer typically presenting as lung
Lymphoepithelioma-like carcinoma - 1 nodules on imaging. Accurate identification of the
benign or malignant nature of these nodules through
Neuroendocrine cancer - 1 imaging features is crucial for clinicians and imaging
Small cell carcinoma - 1 physicians to provide timely and effective treatment for
Total 65 226 patients with malignant lung nodules.

Table 3. Consistency test results of the AI group
AI Pathological findings Total Kappa value P‑value
Malignant nodules Benign nodules
Malignant nodules 204 27 231 0.4883 1.529e-14
Benign nodules 21 35 56
Unidentified 1 3 4
Total 226 65 291
Abbreviation: AI: Artificial intelligence.

Table 4. Consistency test results of the physician‑reading group
Physicians‑reading group Pathological findings Total Kappa value P‑value
Malignant nodules Benign nodules
Malignant nodules 188 22 210 0.5581 2.052e-16
Benign nodules 11 29 40
Calcified nodule 0 1 1
Uncategorized 27 13 40
Total 226 65 291

Volume 3 Issue 3 (2024) 5 doi: 10.36922/td.4178