Health Technology · Supplementary · from Health Technology

Case 1: AI-Assisted Radiology Diagnosis Error

Patient Presentation

Demographics: 54-year-old female office manager

Chief Complaint: "My doctor said the AI found something on my chest X-ray and I need more tests."

History of Present Illness: A 54-year-old woman underwent a routine pre-operative chest X-ray prior to elective cholecystectomy. The radiology department utilizes a commercially available AI-assisted detection system for preliminary reads. The AI algorithm flagged a 1.2 cm opacity in the right upper lobe as "suspicious for malignancy" with a confidence score of 87%. The patient was informed of the finding and referred for CT imaging.

Upon CT chest with contrast, no discrete pulmonary nodule was identified. The opacity seen on chest X-ray corresponded to an overlap of the first rib and clavicle creating a pseudo-nodule artifact. The AI system had generated a false-positive result, causing significant patient anxiety and a two-week delay in her elective surgery.

A retrospective review of the AI system's performance over the prior quarter revealed a false-positive rate of 11.3% for pulmonary nodule detection, with bone overlap artifacts being the most common cause (38% of false positives). The attending radiologist had not reviewed the AI-flagged image before the result was communicated to the referring physician.

Past Medical History:

  • Cholelithiasis (symptomatic)
  • Hypertension, well-controlled
  • Anxiety disorder

Medications:

  • Lisinopril 10 mg daily
  • Sertraline 50 mg daily

Social History:

  • Never smoker
  • Social alcohol use
  • Works as office manager, sedentary lifestyle

Family History:

  • Mother: breast cancer at age 67
  • No family history of lung cancer

Physical Examination

  • Vital Signs: BP 138/86 mmHg, HR 88 bpm, RR 16/min, Temp 36.8°C, SpO2 99% on room air
  • General: Anxious-appearing female, tearful during examination
  • Lungs: Clear to auscultation bilaterally, no wheezes or crackles
  • Cardiovascular: Regular rate and rhythm, no murmurs
  • Abdomen: Mild RUQ tenderness, positive Murphy's sign

Workup and Results

Laboratory Studies:

TestResultReference Range
WBC7,200/µL4,500-11,000/µL
Hemoglobin13.8 g/dL12.0-16.0 g/dL
CEA1.2 ng/mL<3.0 ng/mL
ALT42 U/L7-56 U/L
AST38 U/L10-40 U/L

Imaging/Additional Studies:

  • Chest X-ray (AI-flagged): 1.2 cm opacity in right upper lobe, AI confidence score 87% for malignancy
  • CT Chest with contrast: No pulmonary nodule identified; opacity on X-ray corresponds to first rib-clavicle overlap artifact; lungs clear bilaterally
  • AI System Audit: False-positive rate 11.3% over prior quarter; bone overlap artifacts accounted for 38% of false positives

Clinical Image

Diagram illustrating how overlapping bone structures on chest X-ray can create pseudo-nodule artifacts that trigger false-positive AI detection alerts. Source: Educational illustration.

Diagnosis

AI-Generated False-Positive Pulmonary Nodule Detection Due to Bone Overlap Artifact

Key Diagnostic Criteria:

  • AI system flagged opacity with high confidence score
  • CT chest confirmed no true pulmonary nodule
  • Opacity corresponded to anatomical bone overlap (first rib and clavicle)
  • Workflow gap: AI result communicated before radiologist review

Treatment Plan

  1. Reassurance and patient education regarding the false-positive finding
  2. No pulmonary follow-up required
  3. Proceed with elective cholecystectomy as originally planned
  4. Implement mandatory radiologist verification before AI results are communicated to referring physicians
  5. Retrain AI model with augmented dataset including bone overlap artifacts
  6. Establish departmental protocol for AI-assisted reads requiring attending sign-off

Key Learning Points

  • AI diagnostic tools are adjuncts to, not replacements for, physician clinical judgment
  • False-positive results from AI systems can cause significant patient harm through anxiety, unnecessary testing, and treatment delays
  • Bone overlap artifacts on chest X-ray are a well-known source of false-positive AI detections
  • Institutional protocols must mandate physician review of AI-flagged findings before results reach patients or referring providers
  • Regular performance audits of AI diagnostic systems are essential for quality assurance

All cases for this lecture as Markdown