The Cost-Per-Case Math That Keeps Small and Mid-Size Labs Out of Digital Pathology

# The Cost-Per-Case Math That Keeps Small and Mid-Size Labs Out of Digital Pathology

Ask why a 20,000-case-a-year community lab hasn't gone digital and you will usually be told a story about confidence: pathologists don't trust the screen, the validation burden is scary, the technology isn't ready.

That story is now mostly wrong. Whole slide imaging systems have been cleared for primary diagnosis in the US since 2017, validation methodology is published and stable, and most pathologists under 50 have read digitally at some point in training. The confidence gap has narrowed a great deal.

What has not narrowed is the arithmetic. Digital pathology is a fixed-cost-heavy business, and nearly every commercial and infrastructure model in the market was designed around the volumes of a large academic centre. When you spread those fixed costs across 15,000 cases instead of 250,000, the per-case number roughly triples — and it does so without the small lab getting anything less capable, or anything more.

This article breaks that cost stack down line by line, with real ranges, stated assumptions, and no black-box ROI figure at the end. Every number below is either sourced or explicitly labelled as a modelling assumption you should replace with your own. All figures are USD.

[Image: Hero image — a small community pathology lab bench with a single slide scanner, contrasted against a large centralised scanning array]

The adoption gap is an economics problem sitting on top of a workforce problem

The reason this matters is that the labs least able to absorb fixed costs are the ones with the worst pathologist-coverage position.

The most widely cited workforce analysis of the last decade — Metter et al., *JAMA Network Open*, 2019 — found that the number of active pathologists in the United States fell **17.53%** between 2007 and 2017, while the average case burden per pathologist rose **41.73%** over the same period. That squeeze has not been distributed evenly. Subspecialty depth concentrates in academic and large reference settings; community and regional labs are the ones covering broad case mixes with thin benches and limited backup for vacations, illness, and retirement.

Digital pathology is the most direct available answer to exactly that problem. It is what makes a second opinion a five-minute link instead of a two-day courier round trip, and what makes a part-time remote subspecialist practical. The segment with the sharpest need is the segment the pricing models fit worst.

[Image: Line chart showing US pathologist headcount decline versus case burden per pathologist, 2007–2017, sourced to Metter et al. 2019]

The cost stack, line by line

Here is the model lab used throughout: **25,000 accessioned cases per year, 2.5 stained slides per case, so 62,500 slides per year.** Slides per case varies enormously by specialty mix — 1.5 for a dermatopathology-heavy book, 5+ for GI and prostate biopsy volume. Substitute your own.

Scanner capital

Whole slide scanners are quoted, not listed, and vendor quotes are confidential, so treat these as an **editorial range from market experience, not a published price list**: roughly **$75,000–$150,000** for a low-to-mid throughput scanner (up to ~120 slide capacity), and **$150,000–$300,000** for a high-throughput, high-capacity clinical instrument. Service and maintenance contracts typically run **8–12% of capital cost per year** (modelling assumption).

Check throughput before you check price. At a realistic **60–100 slides per hour** including loading, focus checks, and rescans, our model lab's 62,500 slides spread over 250 working days is **250 slides per day — about 2.5 to 4 hours of scanner time.**

One scanner covers the volume. One scanner also means that when it goes down, you are back on glass with no digital path at all. The redundancy decision is where SMB capital math gets ugly: a second instrument doubles capex to serve a duty cycle you are already only using 40% of. A 250,000-case network buys its fifth scanner and gets redundancy for free.

Assume a **5-year depreciation life** (modelling assumption; useful life is often longer, but clinical support windows and image-format currency tend to govern).

Storage — the line most models get wrong

Start from first principles rather than a vendor's number.

A 15 × 15 mm tissue area scanned at 40× (0.25 µm/pixel) is 60,000 × 60,000 pixels — **3.6 gigapixels**, or about **10.8 GB uncompressed** at 3 bytes per pixel. Apply the JPEG or JPEG2000 compression these systems ship with (commonly **15:1 to 30:1** for brightfield) and you land at **0.36–0.72 GB**. Then add the image pyramid — the stack of downsampled zoom layers — which adds approximately **33%** to the base layer, plus label and macro images.

Real-world observed averages land in a wide band, and the band is the point:

| Slide type | Typical file size | |---|---| | Brightfield, 20×, moderate tissue area | **0.3–0.8 GB** | | Brightfield, 40×, standard surgical | **0.8–3.0 GB** | | Large/whole-mount sections | **3–8 GB** | | Cytology with z-stacking (10–20 focal planes) | **5–20 GB** |

**Your magnification and compression policy is a bigger cost lever than your choice of storage vendor.** Scanning everything at 40× when 20× is diagnostically sufficient for most of your book multiplies your storage bill roughly fourfold, forever.

Using **1.5 GB per slide** as the model midpoint:

- **3.75 GB per case** (2.5 slides × 1.5 GB) - **93,750 GB — 93.75 TB — per year** for our 25,000-case lab - At the low end (0.5 GB/slide): **31 TB/year**. At the high end (3 GB/slide): **188 TB/year** - Over a **10-year retention period**, matching the College of American Pathologists' 10-year recommendation for glass slides: **937 TB, or roughly 0.94 petabytes**

Now price it. Major cloud object storage list rates, US regions, as published (verify current rates — these change):

| Tier | Approx. list price /GB/month | Annual /GB | Retrieval latency | |---|---|---|---| | Standard / hot | **$0.021–$0.023** | ~$0.26 | Milliseconds | | Infrequent access / cool | **$0.010–$0.0125** | ~$0.13 | Milliseconds | | Archive, instant retrieval | **~$0.004** | ~$0.048 | Milliseconds | | Deep archive | **~$0.001** | ~$0.012 | Hours |

Keep all 0.94 PB on hot storage and the tenth-year bill alone is **93,750 GB × 10 years × $0.276 = roughly $259,000 for that year**. Over the full ten years the cumulative spend is about **$1.42 million** — approximately **$9.80 per case in lifetime storage cost**.

Move each cohort to instant-retrieval archive after its first year and the same decade costs roughly **$332,000** — about **$2.15 per case**. Same slides, same accessibility, **4.5× less money.**

Two cautions on that. First, deep archive is not a clinical tier: hours-long retrieval is incompatible with a consult request or a tumour board. Second, archive tiers charge retrieval — around **$0.03/GB**, so roughly **$0.045 to pull back a single 1.5 GB slide.** Negligible per event, material if your recall rate is high; model it at your actual historical slide-retrieval rate rather than assuming it away.

Note the structural point: **storage is the one major line item that is genuinely volume-neutral.** Cost per case is the same at 10,000 cases as at 250,000. Every other line is not.

[Image: Stacked area chart showing cumulative storage volume in TB over 10 years for a 25,000-case lab, with three lines for 0.5 GB, 1.5 GB and 3.0 GB per slide]

Network

93.75 TB per year is **257 GB per day**. What that means depends entirely on your upload capacity:

| Effective upload throughput | Time to move one day's slides | |---|---| | 30 Mbps (typical asymmetric business broadband) | **~19 hours** | | 100 Mbps | **~5.7 hours** | | 500 Mbps | **~1.1 hours** | | 1 Gbps | **~34 minutes** |

This is the line item that ambushes small labs. A lab on a standard business broadband package with 20–50 Mbps upstream cannot digitise a full workday inside a workday. Symmetric fibre at 1 Gbps runs roughly **$500–$2,000/month** and 10 Gbps roughly **$2,000–$8,000/month** (editorial range — circuit pricing is intensely local; get quotes). Clinical continuity argues for a second, diverse-path circuit, so budget **$12,000–$60,000/year** all in.

Egress for viewing is usually smaller than people fear, because tile-based viewers stream only the regions actually examined — call it **50–200 MB per slide viewed** (modelling assumption). At 62,500 views and $0.09/GB, that is roughly **$700–$1,100/year.** Repeat views, tumour boards, and external sharing multiply it.

IT headcount

A digital pathology deployment needs someone who owns scanner uptime, image-management-system administration, storage lifecycle, integration monitoring, and the security posture that a clinical imaging archive demands.

Realistically that is **0.25–1.0 FTE** for an SMB lab, and **not zero** — the failure mode of pretending it is zero is a histotech absorbing it badly on top of a full job. At a fully loaded cost of **$85,000–$130,000** for an imaging/clinical IT specialist (editorial range from public salary data; verify for your market), 0.5 FTE is roughly **$55,000/year**.

At 25,000 cases that is **$2.20 per case.** At 250,000 cases, with 2.5 FTE, it is **$1.10.** The person costs the same; the denominator does not.

Validation

The CAP guideline for validating whole slide imaging for diagnostic use (Pantanowitz et al., 2013; updated by Evans et al., 2022) recommends a validation set of **at least 60 routine cases** for the intended diagnostic use, with a washout period of at least two weeks between glass and digital reads.

Cost that out: 60 cases × 2 reads × 8–15 minutes is **16–30 hours of pathologist reading time**, plus study design, concordance analysis, documentation, and medical director sign-off. A realistic total is **60–120 pathologist-hours**. At a fully loaded pathologist cost of **$175–$225/hour** (derived assumption: ~$350K–$430K fully loaded ÷ 2,000 hours), that is **$11,000–$27,000 of one-time cost** — more if you validate multiple intended uses separately, which most labs should.

Training and the productivity dip

Direct training is the small part: **8–16 hours per pathologist**, **4–8 hours per histotechnologist**. For six pathologists and ten histotechs, roughly **$14,000–$20,000**.

The part nobody budgets is the ramp. Expect pathologists to read **10–20% slower for the first 4–8 weeks** (modelling assumption — track your own, it is the single most useful number you will generate). Six pathologists at 15% for six weeks is **5.4 lost pathologist-weeks**, or roughly **$36,000** of capacity at the rates above.

Combined year-one one-time cost — validation, training, ramp, plus LIS/LIMS integration work at **$15,000–$60,000** — lands between **$75,000 and $160,000**. For a 10,000-case lab, that is **$7.50–$16.00 per case in year one alone**, before a single recurring cost.

[Image: Horizontal bar chart of the one-time year-one cost stack — validation, training, productivity ramp, integration — with low and high estimates per bar]

Putting it together: cost per case at four volumes

Steady state, year three, using the midpoint assumptions above. The platform licence line uses an **illustrative $75,000 annual floor** — a modelling placeholder chosen to represent how minimum-commitment structures behave, not any vendor's published price.

| Line item | 10,000 cases | 25,000 cases | 50,000 cases | 250,000 cases | |---|---|---|---|---| | Scanner (5-yr amortised) | $1.80 | $1.20 | $0.88 | $0.56 | | Service contract | $0.90 | $0.60 | $0.44 | $0.28 | | Storage (10-yr, tiered) | $2.15 | $2.15 | $2.15 | $2.15 | | Network | $1.20 | $0.72 | $0.48 | $0.24 | | IT headcount | $2.75 | $2.20 | $1.65 | $1.10 | | **Infrastructure subtotal** | **$8.80** | **$6.87** | **$5.60** | **$4.33** | | Platform licence (illustrative) | $7.50 | $3.00 | $1.50 | $0.30 | | **Total per case** | **$16.30** | **$9.87** | **$7.10** | **$4.63** |

Two conclusions fall straight out of that table.

**First, the total gap is about 3.5×.** The 10,000-case lab pays $16.30 per case for capability the 250,000-case network gets for $4.63.

**Second — and this is the actionable part — the gap splits almost evenly between two causes.** Infrastructure fixed costs alone account for a **2.0×** spread ($8.80 vs $4.33). That portion is real physics and real payroll. The remaining ~1.5× is contributed entirely by **pricing structure** — floors, minimum commitments, and per-site licences that do not flex with volume.

Roughly half the SMB cost penalty is not a cost at all. It is a commercial model.

[Image: Grouped bar chart comparing cost per case across the four lab volumes, with each bar split into infrastructure subtotal and platform licence]

Why volume-insensitive pricing penalises the labs with the worst coverage problem

A pricing floor is a rational instrument for a vendor: implementation, support, and account management cost roughly the same regardless of customer size, and a floor protects against unprofitable accounts.

The consequence, though, is that the effective per-case price becomes a function of the customer's size rather than the value delivered. A lab at 40% of the floor's break-even volume pays 2.5× the per-case rate for identical software. And because low-volume labs are disproportionately rural, community, and independent — exactly the settings carrying the workforce pressure documented above — the pricing structure systematically prices out the segment with the most acute clinical need.

The counter-argument is that these labs are simply not economical to serve. That is true under a deployment model built around per-site infrastructure, bespoke integration, and high-touch onboarding. It is much less true under shared infrastructure and standardised, self-serve integration. The economics are a function of the delivery model, not a law of nature.

What actually changes the math

Four levers, in descending order of impact for a sub-50,000-case lab.

**Storage tiering, applied from day one.** The single largest controllable line. Automated lifecycle policies moving cohorts from hot to instant-retrieval archive on an age schedule cut ten-year storage cost by roughly **4.5×** — $9.80 to $2.15 per case in our model — with no change to clinical accessibility. Pair it with a scanning-resolution policy: 20× where diagnostically sufficient, 40× where it is not, z-stacking only where cytology demands it.

**Shared infrastructure across a network or consortium.** The scanner duty-cycle problem, the IT FTE problem, and the platform floor problem all dissolve into the same solution: aggregate the denominator. Three 15,000-case labs sharing one image management deployment and one 0.75 FTE administrator move from the left column of that table toward the right, without merging operations or governance. Regional consortium models and hospital-network hub arrangements do this already; the barrier is usually procurement and data-governance structure, not technology.

**Staged rollout instead of full-lab conversion.** Full conversion front-loads every one-time cost simultaneously and maximises the productivity dip. Digitising one workflow first — consults and second opinions, or a single high-volume subspecialty — validates one intended use, trains a subset of the bench, and generates real internal throughput data before the second tranche of capital is committed. It also produces the operational evidence your finance committee will demand for tranche two.

**Consumption-based commercial terms.** If per-case or per-slide pricing without a floor is available, the entire 1.5× pricing penalty in the table above disappears. It is worth asking for explicitly, and worth modelling both ways in your business case.

[Image: Diagram showing a staged rollout sequence — consults, then one subspecialty, then full conversion — with cumulative cost and validated intended uses at each stage]

Run it on your own numbers

You need eleven inputs. Ten of them you already have.

1. Annual accessioned cases 2. Average stained slides per case (from your histology QC data — do not guess this one) 3. Average slide file size in GB (ask for a sample scan of *your* tissue at *your* intended resolution; do not accept a vendor average) 4. Retention period required by your jurisdiction and accreditation body 5. Scanner quote and expected useful life 6. Service contract as % of capex 7. Current upload bandwidth, measured — not the number on the contract 8. Fully loaded cost of the IT FTE fraction you will actually assign 9. Fully loaded pathologist hourly cost 10. Number of pathologists and histotechs requiring training 11. Platform pricing structure — specifically, whether there is a floor and where it sits

Multiply, divide, and you have a defensible cost per case. Then run it twice more: once at your projected volume in three years, and once with tiering and a staged rollout applied. The spread between those three numbers is your negotiating position.

What this model deliberately leaves out

This is a cost model, not an ROI model, and we are not going to hand you a single savings figure derived from assumptions you cannot inspect. The benefit side is real but highly lab-specific, and it includes at minimum:

- **Courier and glass-handling costs avoided** on external consults (roughly **$25–$75 per shipment**, plus staff time, plus the risk of loss in transit) - **Slide retrieval labour** for archived cases — measured in minutes per retrieval and hundreds of retrievals per year - **Turnaround-time improvements** and their downstream effect on referring-clinician retention - **Subspecialty access** without a full-time subspecialty hire — often the largest single value driver for a community lab, and the hardest to put a number on - **Recut and re-stain reduction**, where digital review avoids a physical re-request - **Coverage resilience** — the value of a remote pathologist being able to sign out during a vacancy

Model those against your own baselines. Any vendor who quotes you a headline ROI percentage without asking for those baselines first is quoting you a marketing number.

Where SlidePath sits

We build for the volume band this article describes, and the reason is the arithmetic above rather than sentiment about underserved markets. A delivery model based on shared cloud infrastructure, automated storage lifecycle management, and standardised integration changes which side of that table a 20,000-case lab sits on. We publish our assumptions for the same reason we are publishing this model: finance-side readers should be able to reconstruct any number we give them, and reject it if it does not hold.

If you want to pressure-test your own version of this model against our assumptions, we will walk through it with your figures.

Assumptions and sources

**Published sources**

- Pathologist workforce decline of 17.53% and case-burden increase of 41.73% (US, 2007–2017): Metter DM, Colgan TJ, Leung ST, Timmons CF, Park JY. "Trends in the US and Canadian Pathologist Workforces From 2007 to 2017." *JAMA Network Open*. 2019;2(5):e194337. - Minimum 60-case validation set and washout period for WSI diagnostic validation: College of American Pathologists guideline, Pantanowitz L et al., *Arch Pathol Lab Med*, 2013; updated Evans AJ et al., *Arch Pathol Lab Med*, 2022. - Ten-year glass slide retention recommendation: College of American Pathologists laboratory accreditation retention requirements. - Cloud storage tier pricing: published list rates for major cloud object storage services, US regions. Rates are region-, tier-, and volume-dependent and change; verify against current published pricing before relying on them.

**SlidePath modelling assumptions (replace with your own)**

- 2.5 stained slides per accessioned case; 1.5 GB average slide file size; 250 working days per year - 5-year scanner depreciation; service contracts at 10% of capital cost annually - Scanner capital ranges, network circuit pricing, and IT salary ranges are editorial estimates from market experience, not published price lists - Fully loaded pathologist cost of $175–$225/hour, derived from public compensation survey ranges divided by 2,000 annual hours - 10–20% pathologist productivity reduction for 4–8 weeks post-deployment - 50–200 MB egress per slide viewed - The $75,000 annual platform licence floor is an illustrative modelling placeholder used to demonstrate the behaviour of minimum-commitment pricing structures. It is not any vendor's price, ours or anyone else's.

All figures USD. This model describes infrastructure and operating economics only; it makes no claim regarding diagnostic performance, clinical validation, or regulatory status of any product.