nabu.science
MethodologyAboutContact

 

Reliability concern: This paper was retracted on 2023-12-06. See notice: 10.1155/2023/9796232.

  • Snapshot
  • Key Claims
  • Strengths & Limitations
  • Commentary
  • From the Authors
  • Author Note / Field Feedback
  • Dimension breakdown
  • Impact Profile
Free · About 5 minutes
Report #NBU-F818BC55·Framework v5.0·Pipeline v0.4·Evaluated 01 Jun 2026·OECD: 3.2 Clinical medicine·Subfield: Atomic and Molecular Physics, and Optics·Basis: Full text·Type: Machine learning or AI study

Application of Deep Learning in Lung Cancer Imaging Diagnosis.

Jiang et al.Jiang, Zeng, Wang et al.·1993·Journal of healthcare engineering·10.1155/2022/6107940 ↗

In brief

A competent lung nodule CAD pipeline with competitive LUNA16 benchmark results, undermined by incomplete methodological reporting and absent clinical validation.

ReliabilityConcerns4 of 4 checks completeRead the 3 entries ↓
Abstract

Lung cancer is one of the malignant tumors with the highest fatality rate and nearest to our lives. It poses a great threat to human health and it mainly occurs in smokers. In our country, with the acceleration of industrialization, environmental pollution, and population aging, the cancer burden of lung cancer is increasing day by day. In the diagnosis of lung cancer, Computed Tomography (CT) images are a fairly common visualization tool. CT images visualize all tissues based on the absorption of X-rays. The diseased parts of the lung are collectively referred to as pulmonary nodules, the shape of nodules is different, and the risk of cancer will vary with the shape of nodules. Computer-aided diagnosis (CAD) is a very suitable method to solve this problem because the computer vision model can quickly scan every part of the CT image of the same quality for analysis and will not be affected by fatigue and emotion. The latest advances in deep learning enable computer vision models to help doctors diagnose various diseases, and in some cases, models have shown greater competitiveness than doctors. Based on the opportunity of technological development, the application of computer vision in medical imaging diagnosis of diseases has important research significance and value. In this paper, we have used a deep learning-based model on CT images of lung cancer and verified its effectiveness in the timely and accurate prediction of lungs disease. The proposed model has three parts: (i) detection of lung nodules, (ii) False Positive Reduction of the detected nodules to filter out "false nodules," and (iii) classification of benign and malignant lung nodules. Furthermore, different network structures and loss functions were designed and realized at different stages. Additionally, to fine-tune the proposed deep learning-based mode and improve its accuracy in the detection Lung Nodule Detection, Noudule-Net, which is a detection network structure that combines U-Net and RPN, is proposed. Experimental observations have verified that the proposed scheme has exceptionally improved the expected accuracy and precision ratio of the underlined disease.

Strengths · Limitations

Strengths

#contribution

Competitive LUNA16 benchmark performance across two stages

The pipeline achieves FROC values of 0.876 for nodule detection and 0.883 for false positive reduction on the LUNA16 benchmark, placing it among competitive entries. The detection stage shows particularly high sensitivity, which is clinically valuable for screening applications where recall is prioritized.

↳ Sections 4.1, 4.3; Figures 7-8

#methodological rigour

Systematic comparison of loss functions and network structures

The paper provides internal comparisons of different network structures (ResNet, DenseNet, U-Net variants) and loss functions (cross-entropy vs. weighted cross-entropy) for both false positive reduction and classification stages, with results reported in Tables 1-2. This provides evidence for design choices within the pipeline.

↳ Tables 1-2; Sections 3.3.2, 4.4

#application pathway

Coherent multi-stage pipeline addressing a documented clinical need

The three-stage design — detection, false positive filtering, and classification — mirrors the clinical diagnostic workflow and addresses the documented 30% nodule miss rate by radiologists. The Leaky Noise-or aggregation for multi-nodule classification adds a methodologically distinct element to the classification stage.

↳ Section 1 (Introduction); Section 4.4

Limitations

#methodological rigour

Unconfirmed patient-level data splitting and unclear sample provenance

The nodule detection experiment describes 2000 original samples split 8:2, but does not confirm that splitting was performed at the patient level — a critical requirement in medical imaging to prevent data leakage. The relationship between these 2000 samples and the LUNA16 dataset (888 scans) is unexplained.

↳ Section 4.1

#reporting

Material scope mismatch between Conclusion and paper content

The Conclusion states the paper examines 'CT images of lung cancer and thyroid cancer' and mentions 'benign and malignant classification of the thyroid,' but no thyroid-related methods, data, or results appear anywhere in the paper. This suggests incomplete assembly or revision of a broader work and undermines confidence in the paper's editorial integrity.

↳ Section 5 (Conclusion)

#contribution

Classification results lack external benchmarking

While detection and FPR stages are benchmarked against LUNA16 competition entries, the benign/malignant classification results (Tables 1-2) are compared only internally across the paper's own methods, with no comparison to published state-of-the-art classification approaches. This prevents assessment of whether the classification performance is competitive.

↳ Section 4.4; Tables 1-2

Does this match your read? Add your assessment ↓

Withdrawn — shown for reference only

Nabu Commentary

This paper presents a three-stage lung nodule CAD pipeline with competitive benchmark results on LUNA16 for detection and false positive reduction. The core technical approach — combining U-Net and RPN for detection and introducing Leaky Noise-or for multi-nodule classification — is sound in conception, and the benchmark comparisons provide meaningful external validation for two of the three pipeline stages. However, the paper has notable gaps in methodological reporting that limit confidence: the 2000-sample training set for detection has unclear provenance relative to the 888-scan LUNA16 dataset, patient-level data splitting is not confirmed, and no confidence intervals are reported. The classification stage is evaluated only through internal comparisons without external benchmarking. The mention of thyroid cancer classification in the Conclusion with no supporting content anywhere in the paper is a material inconsistency that suggests the paper may be an incompletely revised version of a larger work, and the duplicate citation number ([5] for both Murphy et al. and U-Net) further points to editorial assembly issues.

Evaluate another paper

Free · About 5 minutes · Research preview

Author Note / Field Feedback

Nabu’s assessment, alongside the field’s view.

Are you an author of this paper?

Dimension breakdown

Evidence Score

Out of 5
012345
2.7±0.2

Contribution
2.9

The three-stage pipeline combining Nodule-Net (U-Net + RPN) for detection, 3D U-Net for false positive reduction, and Leaky Noise-or for classification represents an incremental architectural combination of established components rather than a novel advance. LUNA16 benchmarking provides partial external validation for detection and FPR stages, but the classification stage lacks comparison to published state-of-the-art methods.

Sections 3.2.1, 4.1, 4.4; Tables 1-2; Figures 7-8

second only to 3DCNN_NDET

Methodological Rigour
2.5

The architecture is described in reasonable detail for the detection component and LUNA16 benchmarking is appropriate, but key methodological gaps reduce confidence: patient-level data splitting is not confirmed (Section 4.1), provenance of the 2000-sample training set is unclear, no confidence intervals are reported, no code is released, and the classification stage relies on internal comparisons only. Methods text contains ambiguities and transcription errors.

Sections 3.2-4.4; Data Availability Statement

divided into the training set and test set according to the ratio of 8 : 2

Reporting
2.9

The overall three-stage pipeline structure is logical and followable, but figure cross-reference errors, a duplicate citation number ([5] for both Murphy et al. and U-Net), and the unexplained thyroid cancer mention in the Conclusion reduce navigability and coherence. Writing quality suggests translation from another language with some repetitive passages.

Section 5 (Conclusion); Sections 2 and 3.2.1 (citation [5]); Sections 3.2.2 and 3.2.4 (figure references)

the application of deep learning technology on CT images of lung cancer and thyroid cancer was thoroughly examined

Positioning
2.8

The Related Work section provides adequate coverage of CAD development and LUNA16 competitors, and the paper appropriately positions itself within the benchmark context. However, limitations discussion is effectively absent, with no systematic treatment of failure modes, deployment considerations, generalisability to different scanner types, or clinical translation requirements.

Section 2 (Related Work); Section 5 (Conclusion)

Reliability

Concerns4 of 4 checks


Internal CoherenceNoted

The paper contains a material scope mismatch: the Conclusion (Section 5) mentions thyroid cancer classification, but no methods, data, or results for thyroid cancer appear anywhere in the paper. Additional issues include a duplicate citation number ([5] used for both Murphy et al. and U-Net) and inconsistent figure cross-references.

Research ConductNoted

Data availability is claimed as 'upon reasonable request' with no named repository or persistent identifier, making verification impossible. No ethics approval statement is present, no conflict of interest disclosure is provided, and no code is released. The use of publicly available datasets (LUNA16, LIDC-IDRI) partially mitigates data provenance concerns, but the '2000 original samples' used for nodule detection are not clearly sourced.

Flags: 1 declared / 5 total

Reference IntegrityClean

All 16 cited references resolved to real publications via PubMed, Crossref, or OpenAlex.

References not verified automatically are resolved manually within 24 hours.

Post-Publication RecordConcern

This paper was retracted on 2023-12-06. See notice: 10.1155/2023/9796232.

Last checked: 17 Oct 2026

Sources: Crossref (incl. Retraction Watch) ✓

Impact Profile

Impact areaHealth · Cancer and neoplasms · Tracheal, bronchus, and lung cancer
Type of advanceMethodological
Route to useNamed
AdvanceExtends
BeneficiariesLarge
CitationsFWCI 0.0
Policy · patentNot yet traced
ReachedNot reachedNot assessable or unverified
ReadinessPiloted
Foundational
Controlled
Piloted
Validated
Implementable
FoundationalImplementable

The model is shown working on the standard LUNA16 benchmark test set, but no independent clinical or external cohort validation is reported.

From the paper
“we have developed an intelligent medical detection algorithm to assist doctors in the detection of known organs and diseases”

The paper names doctors as the actors who would use the algorithm to assist detection and improve diagnosis efficiency.

Pathway

Radiologists and hospital imaging services could deploy the deep-learning CAD system as a second reader to detect and classify lung nodules on CT.

Barrier

The model has only been tested on benchmark data, so independent clinical validation and regulatory approval are needed before routine hospital use.

Evaluated 13 Nov 2026 · GBD 2023 · glm_v1_3_3 · impact_profile_v1_3_3

Report an issue →

AI-generated, human-governed. Something look off? Contact us to request a review.

Keep reading

Related evaluations

Other papers Nabu has evaluated blind, starting with the same field.

  1. Independent interferometric observable confirms black hole shadow in M87

    Lai, Thyagarajan, Wong, et al. · 2026 · Physical sciences

    Primary claim

    We reconstruct the black hole shadow image and unequivocally reproduce a ring-like morphology, with an angular diameter of 42.2 ± 1.5 μas

    Tentative
    Evidence Score
    3.5±0.2
    Reliability
    Caveats
  2. Interstellar Dust Inside and Outside the Heliosphere

    2008 · Physical sciences (astronomy and space science)

    Primary claim

    a gas-to-dust-mass ratio in the LIC which is a factor of > 2 larger than the one derived from astronomical observations

    Tentative
    Evidence Score
    3.5±0.2
    Reliability
    Concerns
  3. Discovery of radio emission from the exoplanet $β$ Pictoris b

    Ceballos, Berger, Cendes · 2026 · Physical Sciences

    Primary claim

    We identify the emission as electron cyclotron maser radiation, which implies a magnetic field of ≳1.25 kG at the planet

    Supported
    Evidence Score
    4.2±0.2
    Reliability
    Caveats

How the rubric works →