Radiology Question Quality Review

Executive Summary

This review covers 200 validated non-gold candidate questions sampled from the Radiology pool of 5,382 items, analyzed across eight shards of 25 questions each. The 12 recent PYQs serve as the quality benchmark.

The overall picture is one of a large, structurally uneven bank. A meaningful minority of questions — perhaps 25–30% of the sampled set — meet or approach PYQ standard: they embed clinical context, require genuine reasoning, and use competitive distractors. The remaining 70–75% range from fixable-with-effort to immediately disableable. The problems are not random; they cluster into five distinct, operationally separable issue categories that each imply a different remediation path.

The most urgent problems are: (1) a pervasive overload of Bloom's-1 recall questions that test rote memorization with no clinical framing; (2) a specific and dangerous subset of questions with factually incorrect answer keys; (3) a structural validity failure in image-dependent questions that have been stripped of their images; (4) a contamination problem from dental/cephalometric and non-radiology content; and (5) a large cluster of radiological sign questions that are conceptually appropriate but cognitively too shallow in their current format.

The Bloom's distribution of the sampled set (55 at level 1, 104 at level 2, 21 at level 3, 20 at level 4) confirms the problem quantitatively: over 79% of sampled questions sit at Bloom's 1–2, while the PYQ set demonstrates that the exam increasingly demands Bloom's 3–4 reasoning. The content team should treat this as a structural imbalance requiring systematic action, not isolated fixes.

Headline numbers from the sampled set:

Estimated keep as-is: ~20–25%
Fixable with targeted edits: ~25–30%
Disable recommended: ~45–50%

What Good Looks Like

The PYQ set provides a clear and consistent quality bar. The best items in that set share four properties that the candidate pool frequently lacks.

Clinical anchoring before the question pivot. The neurocysticercosis question (66c11e16) gives age, symptom, and a specific CT finding (cystic lesion with dot sign) before asking for diagnosis. The pediatric chest X-ray question (e8d11636) names a specific population (neonate) and asks for an exception across a set of conditions that all require genuine knowledge to distinguish. Even the simpler PYQs — banana sign (a1efe41d), spinal TB (e0849c8b) — are brief but unambiguous in their clinical framing.

Competitive, clinically plausible distractors. The contrast nephropathy question (15e4d6cd) offers four contrast types that a candidate might plausibly confuse; the breast imaging question (4ed124c2) asks what USG does not offer over mammography, requiring the candidate to hold the properties of both modalities simultaneously. Distractors are not obviously wrong; they require discrimination.

Negative stems used correctly. The neonatal ground-glass EXCEPT question (e8d11636) and the USG-over-mammography NOT question (4ed124c2) use negative framing to test the boundaries of a concept rather than its center. The stem is unambiguous, the exception is genuinely non-obvious, and the correct answer requires reasoning rather than recognition.

Image-based questions that are genuinely image-dependent. The subdural hematoma question (95a5f2b9) and the lacunar infarct question (56bf1e20) require the candidate to interpret an actual image. The question cannot be answered from the text alone, which is the correct design for image-based items.

What the candidate pool frequently does instead: strips the clinical context, asks "sign X is seen in condition Y," uses "all of the above" as the keyed answer, tests a single memorized number or eponym, or references an image that is no longer attached.

Main Issue Categories

1. Bloom's-1 Recall Overload: Questions That Test Memorization, Not Medicine

Why this pattern is bad

The PG entrance examination — NEET-PG and INICET — has progressively shifted toward applied clinical reasoning. A question that asks "what is the half-life of Tc-99m?" or "which radiation has maximum ionizing power?" tests the same cognitive operation as a Class 12 physics exercise. It does not discriminate between a candidate who understands nuclear medicine and one who has memorized a table. At scale, a bank dominated by these items produces a test that rewards cramming over clinical competence, and it fails to prepare candidates for the reasoning demands of the actual exam.

How it shows up

This is the broadest pattern in the sampled set, appearing in every shard. It manifests as: bare-stem definitions (atomic number, atomic weight, gyromagnetic property); isolated numerical recall (half-life, radiation dose limits, permissible exposure); equipment specifications (filament dimensions, silver bromide in X-ray film); and single-association eponym recall with no clinical framing (steeple sign = croup, double bubble sign = duodenal atresia, rat tail = esophageal carcinoma). The easy + Bloom's-1 co-flag appears on at least 8–10 questions per shard, suggesting this pattern accounts for roughly a third of the full candidate pool.

Example question IDs

aeae9a0d (Tc-99m half-life = 6 hours): Pure numerical recall, no clinical application, no tags, no template membership. Adds nothing at PG level.
ec5aa089 (atomic weight = protons + neutrons): Pre-medical chemistry definition. Not a radiology question in any meaningful sense.
da770e27 (most common radiation-induced cancer = leukemia): Universally known, no scenario, no reasoning required.
a55b0fd0 (atomic number = protons): Same problem as ec5aa089; basic chemistry, not radiology.
80b38cfd (gyromagnetic property of hydrogen → MRI): Single-fact recall, Bloom's-1, no tags.
19df3c88 (permissible radiation exposure = 50 mSv/year): Regulatory number recall with no scenario framing.
33117cc4 (least radiosensitive organ = muscle): Bergonié-Tribondeau law at its most basic.
85383278 (photosensitive material in X-ray film = silver bromide): Equipment trivia, no clinical relevance.
e0a6ff1e (steeple sign = croup): Bloom's-1, easy, universally known first-year fact.
54524ba5 (rat tail appearance = esophageal carcinoma): Bloom's-1, easy, no clinical context.
00e4463c (International Day of Radiology = 8th November): Calendar trivia. No PG exam has ever tested this.
75920376 (X-ray tube filament dimensions): Equipment engineering trivia with zero diagnostic application.
d9768c0c (cobalt-60 emits gamma rays): Bloom's-1, easy, rote physics.
4ddb6545 (maximum ionizing power = alpha rays): Standard preclinical fact, no clinical application.
cd7221db (X-rays not emitted by radioactive isotopes): Bloom's-1, easy, pure recall.
ad8ed7ed (radiation exposure occurs in all except MRI): Trivially known by any medical student.
acf5d807 (most common location of hypertensive ICH = basal ganglia): Better tested in Medicine/Neurology; as a standalone radiology question with no imaging context it adds no value.

Recommended disposition

Disable the entire cluster of pure-physics-definition and equipment-trivia questions (ec5aa089, aeae9a0d, a55b0fd0, 80b38cfd, 85383278, 75920376, d9768c0c, 4ddb6545, cd7221db, ad8ed7ed, 00e4463c, e0a6ff1e, 54524ba5, 33117cc4, 19df3c88). For questions where the underlying concept is genuinely examinable at PG level (e.g., radiation dose thresholds, Tc-99m applications, radiosensitivity), the correct remediation is not to fix the existing item but to commission a new question that embeds the concept in a clinical scenario. The existing bare-recall versions should not be retained alongside upgraded versions.

2. Factually Incorrect Answer Keys

Why this pattern is bad

This is the highest-severity issue in the sampled set. A question with a wrong correct answer does not merely fail to discriminate — it actively miseducates. A candidate who studies from a question bank and internalizes an incorrect fact is worse off than one who never encountered the question. In a subject like Radiology where clinical decisions depend on accurate knowledge of imaging findings, factual errors in the answer key are a patient safety concern at one remove.

How it shows up

The pattern appears in two subtypes. The first is outright factual inversion: the keyed answer contradicts established radiological teaching. The second is answer-key inversion in negatively framed stems, where the "except/not true" logic is applied incorrectly, making the wrong option appear correct. Both subtypes appeared repeatedly across shards and likely represent a broader accuracy problem in the full pool.

Example question IDs

187a0076 (Kerley B lines keyed to mitral regurgitation): Kerley B lines are the hallmark of elevated left atrial pressure and pulmonary venous hypertension, most classically associated with mitral stenosis, not regurgitation. This is a direct factual error that will mislead candidates studying for a topic that appears in nearly every PG exam.
f5f25b69 (oxygen timing in radiotherapy keyed as "just before starting"): The oxygen fixation hypothesis requires oxygen to be present during or within microseconds of irradiation. "Just before" is incorrect. The correct answer is option B in the original item. This is a radiobiology error with direct clinical implications for understanding hypoxic tumor radioresistance.
e5ab5a43 (radiation produces its effect by "charring of nucleoprotein"): The accepted mechanism of radiation injury is ionization leading to free radical — primarily hydroxyl radical — generation causing DNA strand breaks. "Charring of nucleoprotein" is an outdated and incorrect description. The question has a wrong correct answer.
995c2217 (miliary mottling keyed to congestive heart failure): Miliary mottling is classically associated with miliary TB, haematogenous metastases, sarcoidosis, and pneumoconiosis. CHF produces pulmonary oedema patterns — Kerley lines, bat-wing opacification, pleural effusions — not miliary mottling. This question will directly mislead candidates on a high-yield chest radiology topic.
09da6b6e (TEE keyed as useful for vascular ring): TEE is limited for vascular ring assessment, which is better evaluated by CT or MRI. TEE is well-established for sinus venosus ASD, aortic dissection, and prosthetic valve thrombosis — all listed as incorrect options in this question. The answer key appears inverted.
2f43667e (best imaging for pheochromocytoma keyed as MRI abdomen): MIBG scintigraphy is the standard answer for functional localisation of pheochromocytoma, especially for extra-adrenal or metastatic disease, in Indian PG curricula. MRI is preferred for anatomical characterisation of adrenal lesions. The key conflicts with standard teaching.
5c9132a0 (scalloping of sigmoid on barium enema keyed to ulcerative colitis): Scalloping/thumbprinting of the sigmoid is more classically associated with ischaemic colitis or pseudomembranous colitis. UC produces loss of haustration and lead-pipe colon. The answer requires expert verification.
02ca6a12 (OPLL — "MRI is best for diagnosis" framed as a false statement): CT is superior for OPLL detection; MRI is better for cord compression assessment. The internal logic of the stem and the keyed answer are inconsistent with each other and with standard teaching.
0e43ec6d (PET scan utility keyed to mycotic aortic aneurysm as the "useful for" answer): Solitary pulmonary nodules, staging lung cancer, and lymph node involvement by malignancy are all validated PET indications. Mycotic aortic aneurysm is the least-established. The question appears to ask what PET is useful for but marks the least-established option as correct. Likely requires rephrasing as "PET is LEAST useful for."
28732557 (Perthes disease — flattening of femoral head as classic sign): Fragmentation of the femoral head epiphysis is the classic early radiological sign. Flattening is a late finding. The answer is debatable and the question needs a stage qualifier.

Recommended disposition

All questions with confirmed factual errors in the answer key (187a0076, f5f25b69, e5ab5a43, 995c2217) should be disabled immediately pending correction. Questions with probable key inversion or strong ambiguity (09da6b6e, 2f43667e, 0e43ec6d, 5c9132a0, 02ca6a12, 28732557) should be flagged for expert radiologist review before any further use. None of these items should remain in active rotation in their current state. For items where the concept is genuinely high-yield (Kerley lines, oxygen effect in radiotherapy, PET indications), a corrected replacement question should be commissioned rather than attempting to patch the existing item.

3. Image-Dependent Questions Rendered Non-Functional by Missing Images

Why this pattern is bad

A question that says "identify the condition in the image below" or "based on the imaging characteristics shown" and then provides no image is not a question — it is an unanswerable prompt. In a text-only or image-stripped delivery context, the candidate either guesses or, worse, reverse-engineers the answer from the options, which rewards test-taking strategy rather than image interpretation. This is particularly damaging in Radiology, where image interpretation is the core clinical skill being assessed. The PYQ set demonstrates the correct model: questions like 95a5f2b9 (subdural hematoma) and 56bf1e20 (lacunar infarct) are explicitly image-dependent and derive their entire discriminatory value from the image.

How it shows up

Across the sampled shards, at least 10–12 questions were identified as image-dependent without a confirmed image present in the text. The pattern appears in two forms: questions where the stem explicitly references "the image below" or "the following radiograph" with no image present, and questions where the clinical vignette is so sparse that it is only answerable with an image (the vignette was written as a caption, not a standalone stem). Several of these carry PYQ tags, suggesting they were originally valid image-based questions that lost their images during migration or reformatting.

Example question IDs

6c6414b3 ("The following image shows the presence of?" — bladder stone, NEET-PG-2018 PYQ): Completely non-functional as text. The PYQ tag confirms this was originally image-based.
20c81912 (references "Radiograph A/B/C" with no images): Cannot be answered without the visual assets.
38da0a89 ("What is the most likely diagnosis?" — no clinical descriptors, no image): Unanswerable.
2e126ea7 ("Which of the following is shown in the barium study?" — no image): Unanswerable.
b5780a18 ("CNS malformation shown below" — no image): Unanswerable.
81843ad7 ("AP view from a percutaneous trans-hepatic cholangiogram" — no image): Unanswerable.
0b74a77e (wrist X-ray for bone age — no image): Unanswerable.
151fc752 (NCCT hemorrhage laterality — PYQ-tagged, no image): Unanswerable without image.
2ab0a79d ("A hysterosalpingogram showing a congenital Mullerian anomaly is depicted" — no image): Unanswerable.
a35f8db0 ("Based on the imaging characteristics of the focal lesion" — no image): Clinical vignette alone insufficient to distinguish hydatid from abscess.

Recommended disposition

These questions should be disabled from active delivery immediately. They should not be fixed by adding descriptive text as a substitute for the image — that changes the question type and the cognitive demand. The correct remediation is to locate and reattach the original image asset. If the image cannot be recovered, the question should remain disabled. For PYQ-tagged items (6c6414b3, 151fc752), the image recovery effort is highest priority because these questions have demonstrated exam relevance. A systematic audit of all image-referencing questions in the full 5,382-item Radiology pool is warranted; the 10–12 instances identified in this 200-question sample suggest the problem may affect hundreds of questions at pool scale.

4. Radiological Signs Tested as Bare Recall Rather Than Applied Reasoning

Why this pattern is bad

Named radiological signs are legitimate and high-yield content for Indian PG exams. The PYQ set confirms this: the neurocysticercosis dot sign (66c11e16), the banana sign in spina bifida (a1efe41d), and the subdural hematoma crescent (95a5f2b9) are all sign-based questions that appear in recent exams. The problem is not the topic — it is the format. When a question simply asks "sign X is seen in condition Y" with no clinical context, it tests whether the candidate has memorized a table of eponyms, not whether they can apply that knowledge to a patient. The PYQ model consistently embeds the sign in a clinical scenario: a child with seizures, a neonate with a specific presentation, a post-operative patient. The candidate pool frequently strips this context entirely.

How it shows up

This is the second-largest pattern in the sampled set after Bloom's-1 recall overload, and it overlaps with it. The distinction worth preserving operationally is that many sign-based questions are at Bloom's 2 (comprehension/application) rather than Bloom's 1, and some have appropriate difficulty ratings — they are not trivially easy, but they are still decontextualized. These are the "fixable" sign questions, as opposed to the Bloom's-1 sign questions that should be disabled. A secondary problem within this category is factual inaccuracy about sign specificity: some questions conflate signs that belong to different conditions (e.g., string sign in pyloric stenosis vs. Crohn's disease), and at least one question uses an unverifiable or non-standard eponym (mother-in-law sign for meningioma, d284bd49).

Example question IDs

fbb55830 (Lyre sign = carotid body tumor): Factually correct, difficulty 3, Bloom's-4 — this is a sign question done reasonably well. The bare-stem format is acceptable at this difficulty level. Keep; consider adding a brief clinical stem.
b63bed18 (Carman meniscus sign): Bare-stem recall, no clinical context. Fix by adding a vignette.
27b3b5b0 (rib notching in coarctation): Bare-stem recall. Fix by embedding in a clinical scenario (young hypertensive patient, CXR findings).
5324c498 (signet ring sign): Bare-stem, Bloom's-1. Disable or upgrade.
9087cb8e (coeur en sabot in Tetralogy of Fallot): Acceptable as a medium-difficulty cardiovascular radiology item; keep.
b72d31db (molar tooth sign = Joubert syndrome): Bloom's-4, well-differentiated distractors. Keep as-is — this is a sign question at the right cognitive level.
ffd064c3 (string sign with "all of the above" as correct): Conflates pyloric stenosis and Crohn's/TB string signs under one answer. Fix: restructure to ask for the classic association and remove "all of the above."
d284bd49 (mother-in-law sign for meningioma): Non-standard eponym not found in major Indian PG references. The established sign is the dural tail sign. Fix: verify source; if unverifiable, replace with dural tail sign.
18f2442f (target sign = CHPS): Bloom's-1, easy, "all the above" distractor. Disable.
924a556a (bare orbit sign = NF-1): Bloom's-1, easy. Disable or upgrade to a vignette with NF-1 clinical features.

Recommended disposition

This category requires stratified handling rather than a single disposition. Sign questions at Bloom's 3–4 with clinical context or competitive distractors (b72d31db, fbb55830, 9087cb8e, 040454a9) should be kept. Sign questions at Bloom's 2 with correct factual content but no clinical framing (b63bed18, 27b3b5b0, a16a47d5) should be fixed by adding a brief clinical stem — this is a tractable edit. Sign questions at Bloom's 1 that are flagged easy and test a universally known association (18f2442f, 924a556a, 5324c498, 54524ba5) should be disabled. Sign questions with factual accuracy concerns (ffd064c3, d284bd49, 5c9132a0) should be fixed or disabled depending on whether the concept can be salvaged with a rewrite.

5. Structural Question Defects: "All of the Above," Tautological Options, and Broken Distractor Sets

Why this pattern is bad

These are construction-level failures that undermine the psychometric validity of an otherwise potentially correct question. "All of the above" as the keyed answer rewards test-taking strategy (if you know two options are correct, the answer must be "all of the above") rather than knowledge. Tautological options — where the correct answer repeats a term from the question stem — have zero discriminatory value. Broken distractor sets — where one option is a morphological descriptor in a question asking for an imaging sequence, or where one distractor is an obsolete investigation — collapse the question to a two-option choice and inflate apparent difficulty.

How it shows up

"All of the above" as the correct answer appears in at least six questions across the sampled set (3194e0ee, ffd064c3, a0e6fa62, f26f5008, 2ddd3270, 379db4f6). The tautological option appears in cc5a8852 (orthopantomogram question). Broken distractor sets appear in 012d3099 (morphological descriptor as an MRI sequence option), a5ff57f8 (air encephalography as a distractor in a modern imaging question), 1c9bb245 (tomography as a metastasis screening tool), and e383863b (cephalometric trivia with no plausible wrong answers). The "None of the above" variant appears in 271763c7 (bone scintigraphy cannot detect).

Example question IDs

cc5a8852 ("Which investigation is known as an orthopantomogram?" → correct answer: "Orthopantomogram"): The correct answer is identical to the term in the question stem. Zero discriminatory value. This is the most egregious example in the sampled set.
3194e0ee (achalasia cardia radiological features, "all of the above" correct): Rewards test-taking strategy. Fix: restructure as a positive stem asking for the most specific or earliest radiological sign.
f26f5008 (X-ray quality factors, "all of the above" correct): Same structural problem. Fix: rewrite as single best answer or "which does NOT influence" format.
379db4f6 (miliary mottling causes, "all of the above" correct): Factually defensible but structurally weak. Fix: replace with a specific condition as the keyed answer.
012d3099 (arachnoid vs. epidermoid cyst differentiation): Option C is "smooth margin" — a morphological descriptor, not an imaging sequence. The option set is incoherent. Fix: replace with a valid MRI sequence (DWI) and reconsider whether FLAIR or DWI is the correct answer.
2ddd3270 (renal function evaluation, "all of the above" correct): Fix: ask which single modality is best for a specific clinical scenario.
271763c7 (bone scintigraphy cannot detect, "none of the above" correct): Fix: add a genuine exception (e.g., purely lytic myeloma lesions) and re-key.
a0e6fa62 (Doppler USG in twin pregnancies, "all of the above" correct): Doppler is not a primary tool for diagnosing twins; the "all of the above" structure masks this inaccuracy. Fix: remove "diagnosis of twins" as a distractor or restructure.

Recommended disposition

The tautological question (cc5a8852) should be disabled immediately — it cannot be fixed without a complete rewrite that changes the question entirely. "All of the above" questions where the underlying concept is high-yield (3194e0ee, 379db4f6) should be fixed by restructuring the option set; this is a moderate-effort edit. "All of the above" questions where the underlying concept is low-yield or the factual content is also questionable (a0e6fa62, f26f5008) should be disabled. Questions with broken distractor sets (012d3099, a5ff57f8) should be fixed if the concept is worth retaining, disabled if the concept is already well-covered by better questions elsewhere in the pool.

6. Out-of-Scope and Misclassified Content

Why this pattern is bad

Questions that belong to a different subject or a different specialty sub-domain dilute the Radiology pool and create confusion for candidates using topic-filtered study modes. More practically, they consume slots in the question bank that should be occupied by high-quality Radiology content. Misclassification also corrupts difficulty calibration and Bloom's distribution metrics for the subject.

How it shows up

Two distinct subtypes appear in the sampled set. The first is genuine subject contamination: questions that have no radiological content and belong entirely in another subject. The second is dental/cephalometric radiology intrusion: questions on cephalometric projections, orthopantomograms, and dental caries that are relevant to dental PG exams but not to NEET-PG/INICET Radiology. A third minor subtype is topic-label misclassification within Radiology (e.g., a gastric cancer question asking about endoscopy labeled under Abdominal and Pelvic Radiology, or an ECG question labeled under Pediatric Radiology).

Example question IDs

2397438f ("ECG shows which electrolyte abnormality?" labeled Pediatric Radiology): No radiological content whatsoever. This is a Cardiology/Medicine question. Disable from Radiology pool.
e383863b (cephalometric X-ray source-to-subject distance = 5 feet): Dental radiology trivia, Bloom's-1, not relevant to medical PG Radiology. Disable.
e1edfa65 (disadvantage of cephalometric radiograph): Dental/orthodontic scope, outside medical PG Radiology. Disable.
9460a0b4 (cervical burnout as false positive for dental caries): Dental radiology, outside medical PG scope. Disable.
cc5a8852 (orthopantomogram identification): Dental radiology nomenclature. Disable (also tautological, as noted above).
a757b7ab (Gardner's view = two molar view): Dental radiology nomenclature recall, Bloom's-1, easy. Disable.
083a0127 (gastric cancer first-line investigation = endoscopy, labeled Abdominal and Pelvic Radiology): Endoscopy is not a radiology investigation. Reclassify to GI Surgery/Medicine or reframe to ask about the role of staging CT after endoscopic diagnosis.
9f6a8012 (investigation of choice for laryngeal cartilage involvement, labeled Neuroradiology): Should be Head & Neck Radiology or ENT Radiology. Reclassify.

Recommended disposition

The ECG question (2397438f) and all dental/cephalometric questions (e383863b, e1edfa65, 9460a0b4, cc5a8852, a757b7ab) should be disabled from the Radiology pool. If a dental radiology pool exists or is planned, the dental questions could be migrated there; otherwise disable entirely. The gastric cancer question (083a0127) should be reclassified or reframed. The laryngeal cartilage question (9f6a8012) should be reclassified within Radiology. Given that five dental/cephalometric questions appeared in a 200-question sample, the full pool of 5,382 likely contains a proportionally larger contamination cluster that warrants a targeted audit.

7. Investigation-of-Choice Questions: Duplicate Density and Missing Clinical Scaffolding

Why this pattern is bad

"Investigation of choice for condition X" is a legitimate and frequently tested format in Indian PG exams. The problem in the candidate pool is twofold. First, these questions appear in near-duplicate pairs — the same clinical question asked twice with the same answer, inflating the apparent coverage of a topic without adding discriminatory value. Second, even the non-duplicate versions frequently present as bare-stem recall ("IOC for gallstones = USG") rather than as clinical scenarios that require the candidate to reason about why one modality is preferred over another in a specific context. The PYQ benchmark (e.g., 15e4d6cd — contrast agent in renal dysfunction) shows that IOC questions are most valuable when they embed a clinical constraint that makes the choice non-trivial.

How it shows up

The most direct example is the gallstone duplicate pair (c395a786 and 50f11fc1), both asking for the investigation of choice for gallstones, both answering USG, both flagged easy. Across the sampled set, IOC questions without clinical scaffolding appear for biliary pathology (08401876), CBD evaluation (40dc1faf), recurrent GIST (44168eed), pheochromocytoma (2f43667e — also a factual accuracy concern), and renal function (2ddd3270 — also an "all of the above" concern). The pattern is broad enough to suggest that the full pool contains many more such pairs and bare-stem IOC items.

Example question IDs

c395a786 / 50f11fc1 (duplicate pair — IOC for gallstones): One must be disabled. c395a786 is the weaker version (no tags, no template membership). Disable c395a786; retain 50f11fc1 only after upgrading to a clinical scenario.
08401876 (IOC for biliary pathology — bare stem): Fix by adding clinical context (e.g., patient with jaundice and dilated ducts on USG, next investigation).
40dc1faf (IOC for CBD evaluation — bare stem): Same fix path.
44168eed (IOC for recurrent GIST — bare stem): Factually correct (PET-CT for GIST on imatinib). Fix: add a brief clinical stem (patient on imatinib, follow-up imaging).
d49ff758 (EUS for esophageal T-staging): This is an IOC question done well — it specifies the clinical purpose (T-staging) and the modality choice is non-trivial. Keep as a model for the fix path.
801b4892 (radionuclide scan for renal function): Clinically grounded, appropriate. Keep.

Recommended disposition

Disable one item from each confirmed duplicate pair. For bare-stem IOC questions where the underlying concept is high-yield and not already covered by a better question, fix by adding a clinical stem — this is a tractable edit that elevates the question from Bloom's-1 recall to Bloom's-2 or Bloom's-3 application. For bare-stem IOC questions where the concept is low-yield or already well-covered, disable. A deduplication audit of the full Radiology pool specifically targeting IOC questions is recommended, with particular attention to abdominal radiology and nuclear medicine where the duplicate density appears highest in this sample.

Prioritization

The five actions below are ordered by urgency and impact. The first two require immediate action before the next content release cycle; the remainder can be addressed in a structured remediation sprint.

Priority 1 — Immediate disable: Factually incorrect answer keys (Category 2) Questions with confirmed wrong correct answers (187a0076, f5f25b69, e5ab5a43, 995c2217) and probable key inversions (09da6b6e, 0e43ec6d) must be removed from active delivery immediately. These questions are actively harmful. This is a small, identifiable set and the action is unambiguous.

Priority 2 — Immediate disable: Image-dependent questions without images (Category 3) Questions that are unanswerable without a missing image (6c6414b3, 20c81912, 38da0a89, 2e126ea7, b5780a18, 81843ad7, 0b74a77e, 151fc752, 2ab0a79d, a35f8db0) should be removed from active delivery immediately. A full audit of the 5,382-question pool for image-referencing questions is warranted given the density of this problem in the sample.

Priority 3 — Systematic disable: Bloom's-1 recall overload and out-of-scope content (Categories 1 and 6) The large cluster of pure-physics-definition, equipment-trivia, and dental/cephalometric questions should be disabled in a single curation pass. These require no expert review — the disable decision is clear from the question text alone. This action will meaningfully improve the Bloom's distribution of the active pool.

Priority 4 — Targeted fix: Radiological sign questions needing clinical scaffolding (Category 4) Sign questions at Bloom's 2 with correct factual content but no clinical framing represent the largest fixable cluster. Adding a two-sentence clinical stem to each is a tractable edit that can be batched. Priority within this cluster should go to high-frequency exam topics: chest radiology signs, cardiovascular silhouettes, and neuroradiology signs.

Priority 5 — Structural repair: "All of the above" and broken distractor sets (Category 5), and IOC deduplication (Category 7) These require more careful editing — restructuring option sets and verifying that the revised question has a single unambiguous correct answer. Deduplication requires a cross-pool comparison. These are medium-effort tasks that should follow the higher-priority disables.

Example Keep / Fix / Disable Calls

The following table summarizes representative disposition calls drawn from the sampled set. These are illustrative of the patterns described above, not an exhaustive list.

KEEP

Question ID	Topic	Reason
dac48357	Neuroradiology	Well-constructed clinical vignette (pilocytic astrocytoma), age + location + imaging characteristics, unambiguous answer, Bloom's-2 applied
b72d31db	Neuroradiology	Molar tooth sign → Joubert syndrome; Bloom's-4, well-differentiated distractors, appropriate difficulty
040454a9	Chest Radiology	PE signs EXCEPT Monod sign; classic high-yield distinction, appropriate format
6511d9c2	Paediatric Radiology	Drooping water lily sign in 2-year-old with recurrent UTI; PYQ-tagged, Bloom's-4, clinically anchored
af59872f	Neuroradiology	Craniopharyngioma EXCEPT question; PYQ-tagged, Bloom's-4, well-constructed negative stem
736d4030	Contrast Media	Contrast reaction with urticaria/wheezing/tachycardia → epinephrine; clean clinical scenario, Bloom's-3
cfcbc2d7	Abdominal Radiology	Retroperitoneal necrotic mass near left renal hilum in 24-year-old male; Bloom's-3, plausible distractors
ff14c657	Chest Radiology	Galaxy sign + bilateral hilar LAD → sarcoidosis; integrative Bloom's-4 question
4e061105	Chest Radiology	Systemic sclerosis vignette → HRCT; PYQ-tagged, Bloom's-3, competitive distractors
d7e682af	Emergency Radiology	Blunt abdominal trauma, haemodynamic parameters → FAST vs CT decision; Bloom's-3
42d08d90	Contrast Studies	Dense persistent nephrogram EXCEPT; INICET PYQ-tagged, Bloom's-4
927de5a9	Abdominal Radiology	Comb sign → Crohn's disease; clean, unambiguous, clinically relevant
9c087f1f	Chest Radiology	Silhouette sign obliterating left cardiac border → lingula; medium difficulty, good distractors
54ed5052	Chest Radiology	Boerhaave syndrome → Naclerio V sign; clinical vignette, Bloom's-3, defensible answer
1e29fecd	MSK Radiology	Eggshell calcification EXCEPT osteogenic sarcoma; appropriate difficulty, tests applied knowledge
0038cbff	Ultrasound	EUS chronic pancreatitis duct diameter; NEET-PG tagged, difficulty 3, Bloom's-4
d5c34af9	Nuclear Medicine	MUGA scan for anthracycline cardiotoxicity; specific, clinically relevant, applied nuclear medicine

FIX

Question ID	Topic	Problem	Recommended Fix
187a0076	Chest Radiology	Kerley B lines keyed to mitral regurgitation — factual error	Change correct answer to mitral stenosis
f5f25b69	Radiobiology	Oxygen timing in radiotherapy keyed as "just before" — factual error	Re-key to "during and within microseconds of starting"
0e43ec6d	Nuclear Medicine	PET utility — answer key appears inverted	Rephrase as "PET is LEAST useful for" or correct key
2f43667e	Nuclear Medicine	Pheochromocytoma best imaging keyed to MRI — conflicts with MIBG standard	Correct key to MIBG or reframe stem to specify anatomical characterisation
ffd064c3	Abdominal Radiology	String sign — "all of the above" conflates pyloric stenosis and Crohn's	Restructure to ask for classic association; remove "all of the above"
012d3099	Neuroradiology	Arachnoid vs epidermoid — "smooth margin" is a morphological descriptor, not an MRI sequence	Replace with DWI as an option; reconsider whether FLAIR or DWI is keyed
3194e0ee	Abdominal Radiology	Achalasia features — "all of the above" correct	Restructure as positive stem asking for most specific/earliest sign
63d4d172	Neuroradiology	Ependymoma vs aqueductal stenosis — stem needs clarifying detail	Add location of lesion on CT or age-specific clue to unambiguously favour ependymoma
b24fd273	Ultrasound	Probe frequency matching — 20 MHz for linear probe is non-standard	Expert review and correction of frequency values
083a0127	Abdominal Radiology	Gastric cancer IOC = endoscopy — not a radiology investigation	Reclassify or reframe to ask about staging CT role
9259f59b	Radiation Oncology	Melanoma radioresistance — Bloom's-1, no clinical context	Upgrade with clinical stem about palliative radiotherapy decision
44168eed	Abdominal Radiology	IOC for recurrent GIST — bare stem	Add clinical stem: patient on imatinib, follow-up imaging
5f435213	MSK Radiology	NOT a feature of scurvy — keyed answer debatable	Expert review of keyed answer and distractor wording; retire one of the two scurvy questions
de769f91	Chest Radiology	Pulmonary venous hypertension feature — late-stage finding presented without stage qualifier	Add "chronic/long-standing" qualifier or reframe as "late-stage feature"
a2e1637a	Cardiovascular Radiology	ASD vs VSD on CXR — stem too sparse	Add shared features (pulmonary plethora, cardiomegaly) and ask what distinguishes them
30dff6ff	Obstetric Radiology	Placenta praevia vs accreta conflated in stem	Specify "for localisation of placenta praevia" to remove ambiguity

DISABLE

Question ID	Topic	Reason
ec5aa089	Radiation Physics	Atomic weight definition — pre-medical chemistry, not radiology
aeae9a0d	Nuclear Medicine	Tc-99m half-life — rote numerical recall, no clinical application
da770e27	Radiobiology	Most common radiation-induced cancer — universally known, no reasoning required
00e4463c	Radiation Physics	International Day of Radiology date — calendar trivia
75920376	Radiation Physics	X-ray tube filament dimensions — equipment engineering trivia
e5ab5a43	Radiobiology	Radiation mechanism = "charring of nucleoprotein" — factually incorrect answer key
995c2217	Chest Radiology	Miliary mottling keyed to CHF — factually incorrect answer key
2397438f	Pediatric Radiology	ECG and electrolyte abnormality — no radiological content, wrong subject
e383863b	Dental Radiology	Cephalometric source-to-subject distance — dental scope, not medical PG Radiology
e1edfa65	Dental Radiology	Disadvantage of cephalometric radiograph — dental scope
9460a0b4	Dental Radiology	Cervical burnout as false positive for dental caries — dental scope
cc5a8852	Dental Radiology	Orthopantomogram identification — tautological and dental scope
a757b7ab	Dental Radiology	Gardner's view = two molar view — dental nomenclature, Bloom's-1
67e09762	Abdominal Radiology	Graham Cole test — eponym recall, obsolete investigation
066a8ebc	Radiation Physics	Highest penetration power — rote physics, no clinical application
85383278	Radiation Physics	Silver bromide in X-ray film — equipment trivia
c06e9a09	Radiation Physics	Linear accelerator energy type — conceptually incoherent question
6c6414b3	Abdominal Radiology	Image-dependent, no image present, unanswerable as text
38da0a89	Abdominal Radiology	Image-dependent, no image present, unanswerable as text
2e126ea7	Abdominal Radiology	Barium study image question, no image present
0b74a77e	MSK Radiology	Wrist X-ray for bone age, no image present
151fc752	Neuroradiology	NCCT hemorrhage laterality, PYQ-tagged, no image present
c395a786	Abdominal Radiology	Duplicate of 50f11fc1 — IOC for gallstones, weaker version
18f2442f	Abdominal Radiology	Target sign = CHPS — Bloom's-1, easy, "all the above" distractor
e0a6ff1e	Chest Radiology	Steeple sign = croup — Bloom's-1, easy, universally known
91ac1705	Breast Imaging	Mammography as screening — Bloom's-1, easy, non-credible distractors
ffb068d2	Neuroradiology	Myelography dye space — Bloom's-1, easy, pure recall
b3f59d2d	Nuclear Medicine	Phosphorus-32 emits beta particles — Bloom's-1, easy, rote physics
700e7318	Radiation Physics	X-rays cause damage by ionisation — Bloom's-1, easy, no clinical application
1d383812	Paediatric Radiology	Double bubble sign = duodenal atresia — Bloom's-1, easy, universally known
8301fc50	Abdominal Radiology	Commonest incidentaloma in liver = haemangioma — Bloom's-1, easy, no clinical context
ad8ed7ed	Radiation Physics	Radiation exposure in all except MRI — trivially known
cd7221db	Radiation Physics	X-rays not emitted by radioactive isotope — Bloom's-1, easy
4ddb6545	Radiation Physics	Maximum ionizing power = alpha rays — standard preclinical fact
33117cc4	Radiobiology	Least radiosensitive organ = muscle — Bloom's-1, easy
19df3c88	Radiation Safety	Permissible exposure = 50 mSv/year — regulatory number recall, no scenario
a55b0fd0	Radiation Physics	Atomic number = protons — basic chemistry, not radiology
80b38cfd	MRI Physics	Gyromagnetic property → MRI — Bloom's-1, easy, no tags
9d380b4a	Nuclear Medicine	Isotope in RAIU = I-123 — Bloom's-1, easy; disable or reframe as clinical application
b25e6ce2	Abdominal Radiology	Hydatid cyst EXCEPT Mercedes Benz sign — Bloom's-1, easy, factual accuracy concern