The Finding You Almost Missed

There is a particular kind of discomfort a radiologist knows well: the moment on a second read when a finding comes into view that was not in the report. It is not always a critical miss. Sometimes it is a small thing, an incidental nodule noted and forgotten, a subtle periosteal reaction that did not register under the weight of thirty other studies. But sometimes it is not small at all. The question is not whether readers miss findings. The question is whether there is anything systematic that can help.

The weight of the worklist

Radiology caseloads have grown steadily for two decades. Across the NHS, a consultant radiologist may report over fifteen thousand studies in a year, across modalities ranging from plain film to cross-sectional CT, often with PACS open on multiple screens and an RIS queue ticking in the corner. The clinical consequences of delay are well documented: missed incidental pulmonary nodules found on a staging CT, rib fractures misattributed to normal ageing, early apical opacities on a chest film reported as clear. None of these are the result of incompetence. They are the result of a system in which attention is a finite resource and the demand on it is not.

Fatigue and the searching eye

Two cognitive phenomena shape how radiologists miss findings. The first is satisfaction of search, the well-studied tendency to stop looking once the first significant abnormality is found. A reader who correctly identifies a large pneumothorax on a chest film may then underweight a subtle right-base opacity sitting in the same study. The second is inattentional blindness, the failure to notice a stimulus that was not the focus of the search. Neither is a lapse of attention in the ordinary sense. Both are normal features of how visual attention works under load, and both become more pronounced as session length increases and cognitive resources are drawn down. The afternoon read of a busy acute list is not the same as the first study of the morning.

Where findings tend to hide

• Apical and subphrenic lung regions, which sit at the edges of the visual field during a routine chest review.
• Bony structures on a CXR, which are deprioritised when the clinical request is focused on the lung parenchyma.
• Rib fractures on CT without reformats, particularly when the presenting concern is soft tissue.
• Incidental findings on a study requested for a different reason, where the report template anchors attention to the stated clinical question.
• Subtle ground-glass opacities in the lung bases, particularly at mediastinal window widths.
• Small pleural effusions, especially right-sided, in the context of a cardiomegaly report.

What a second reader does

A second read changes the dynamic in one specific way: it is conducted without the anchoring effect of the first. A colleague who reads a study cold, without seeing the primary report, brings fresh attention to the image. This is the well-established rationale for double-reading programmes in breast screening, where two independent reads are standard practice. The challenge is resource. A programme that doubles the reading requirement for every study is not deliverable in a system already under pressure, which is why double-reading in most settings is reserved for high-risk cases or specific modalities. Verityn exists to make the principle available at scale. It reads every study independently, without prior report, without session fatigue, without anchoring to the clinical question, and it flags anything that meets its detection criteria with a location, a heatmap, and a stated confidence.

The second reader does not need to be right every time. It needs to be looking at the right places.

Calibrated, not infallible

A model that flags every study will rapidly lose clinical trust. A model that flags too little will not be worth the workflow step. Verityn is designed to flag with a stated confidence, and that confidence needs to mean what it says. When the system reports a sixty-eight percent confidence on a right lower lobe density, the radiologist should understand this as a genuine expression of uncertainty, not a product of post-hoc scaling. Calibration is the hardest part of building honest AI in a clinical setting, and it is the part that demands the most care.

The radiologist is still the radiologist. Every finding Verityn raises is a prompt for fresh attention, not a second opinion to accept or dismiss. The clinician reads the study, considers the flag, applies their clinical knowledge, and reports. The responsibility does not move. What changes is that a second set of eyes, one that does not tire, has already been through the same image. That is not a small thing.