The British Journal of Radiology, 74 (2001), 317322

2001 The British Institute of Radiology

Geographical distribution of breast cancers on the

mammogram: an interval cancer database
M BROWN, C ECCLES, MSc and M G WALLIS, FRCR
Warwickshire, Solihull and Coventry Breast Screening Service, Coventry and Warwickshire Hospital, Stoney
Stanton Road, Coventry CV1 4FH, UK

Abstract. Auditing interval cancers is an important part of a breast screening radiologists

continuing education. We set out to determine whether the position of interval cancers on the
mammogram differs from those detected at screening. The 773 interval cancers so far identified,
and the first 200 screen detected cancers, have been entered onto a Microsoft Access 97 database
developed to record pathological and radiological features, including the position of the cancer on
a stylized diagram using a point and click system. Reports were generated showing positions of
all interval cancers by classification and reader. The distribution of true interval cancers is
statistically different from screen detected cancers on both views. The distribution of the false
negative and screen detected cancers only differs on the oblique view. False negative and true
interval cancers are of the same distribution on both craniocaudal and oblique views. However,
these differences do not appear to be practically useful when applied to individual readers. We
have developed a database that allows systematic recording of pathological and radiological
information regarding breast cancers. Additionally, it can record the geographic position of the
cancer with minimal memory requirements. Statistical differences in the distribution of false
negative and screen detected cancers have been demonstrated and the stylized diagrams reinforce
the importance of the conventional review areas. Although this has not identified any blind spots
in our own readers, it nevertheless provides film readers with a tool to audit their own work.

From its inception in 1988, the UK National

Health Service Breast Screening Programme
(NHSBSP) recognized that screening can never
be a perfect test and that cancer will present
during the interval between screens [1]. The
number of these interval cancers is a measure of
the quality of the programme and will indirectly
predict the mortality reduction that can be
expected [2, 3]. The NHSBSP expects all film
readers to audit interval cancers as part of their
continued professional development, in the anticipation that retrospective review of those cancers
that could potentially have been diagnosed earlier
will help to improve performance [4]. Several
groups have identified radiological features [57]
that film readers find difficult, but to date there
has only been limited work describing where
cancers occur on the actual mammograms [810].
Whilst developing a database to collect and audit
interval cancers, we have developed a relatively
simple way of plotting the position of the cancer
on a stylized diagram. We can then systematically
Received 11 September 2000 and in revised form 24
November 2000, accepted 5 December 2000.
Address correspondence to Dr MG Wallis.
Melita Brown was funded by the Breast Screening Trust
Fund.
The British Journal of Radiology, April 2001

collect information, not only about radiological

signs but also about geographical distribution of
interval cancers, to assist film readers in their
continual improvement.

Materials and methods

All 773 interval cancers identified up to 1 May
2000 by the Warwickshire, Solihull and Coventry
Breast Screening Programme, since screening
started in 1989, have been classified by both the
internal film readers and as part of Regional
Quality Assurance initiatives in line with
NHSBSP guidance [4]. These have all been
entered onto the bespoke database. In addition,
the first 200 screen detected cancers have been
entered to form a control group.
The bespoke database has been developed using
Visual Basic for Applications in the Microsoft
Access 97 software environment. The database
records the patients demographic details, clinical
information (including pre-operative tests and
results), radiological features of the diagnostic
mammogram and last screening mammogram,
histopathology results and audit details including
reader details and classification. A simple point
and click system is used to mark the position of
the cancer on a stylized diagram of the breast
317

M Brown, C Eccles and M G Wallis

using both mediolateral oblique and craniocaudal

projections. The position is recorded by the
computer as an xy co-ordinate. The stylized
diagram can be displayed and printed. Variables
can be selected to include any or all of the film
readers and interval classification. For convenience, right-sided cancers are transposed onto and
combined with left-sided cancers.
A fine grid is placed over the cancer location
diagram created by the database to determine
whether the geographical positions of cancers
within the breast have a random distribution. The
number of cancers in each square can then be
observed. The Poisson, x2 and critical values are
calculated using this information. It is then
possible to ascertain whether the distribution of
cancers within the breast is random.
A two-sample KolmogorovSmirnov test is
performed to establish whether there is a difference in the distribution of the false negative and
true interval cancers. A marker is placed on both
the oblique and craniocaudal projections using
the diagram showing the geographical location of
the cancers from the developed database. The
distance from the marker to each individual
cancer is measured to the closest millimetre.
From this, the cumulative relative frequency
polygon is developed for each projection and
interval cancer classification. This method is
repeated to compare the geographical distribution
of the screen detected cancers and the two
classifications of interval cancers. By comparing
the maximum difference between the cumulative
relative frequency polygons and the critical value,
it is possible to conclude whether the distributions
are significantly different and at what level.

The distribution of cancers across the breast is

uneven, with clusters in the tail and inframammary fold on the oblique projection. On the
craniocaudal projection, clustering occurs in the
tail, centrally and in the medial portions of the
breast. This is confirmed by using the Poisson
distribution (critical value 15.087 (3 dp)) and the
x2 test (x2 value 527.824 (3 dp)).
Simple visual examination of the figures does

Results
The classification of the 773 interval cancers is
shown in Table 1.
Figure 1 shows the distribution of the 200
screen detected cancers. The distribution of the
354 true interval cancers is shown in Figure 2 and
the 122 false negative interval cancers in Figure 3.
The false negative interval cancers for two of the
readers is shown in Figure 4.
Table 1. The number and classification of all interval
cancers at the Warwickshire, Solihull and Coventry
Breast Screening Service
Classification

Number

True
False negative
Occult
Unclassified
Minimal signs

364
122
107
135
45

47.1
15.8
13.8
17.5
5.8

Total

773

100
Figure 1. Distribution of screen detected cancers.

318

The British Journal of Radiology, April 2001

Interval cancer geography: a database

Figure 2. Distribution of true interval cancers.

not suggest any obvious differences in distribution

between screen detected controls and the true or
false negative interval cancers. However, the
following results, shown in Table 2, have been
obtained using the KolmogorovSmirnov test.
Table 2 shows that the distribution of true
interval cancers is significantly different from the
distribution of screen detected cancers on both
projections. The distributions of false negative
interval cancers and screen detected cancers are
the same on the craniocaudal projection, but are
The British Journal of Radiology, April 2001

Figure 3. Distribution of false negative interval cancers.

of a different distribution on the oblique projection. It can also be seen that false negative and
true interval cancers are of the same distribution
on both the craniocaudal and oblique projections.

Discussion
Our frequency of false negative cancers, 19% of
classifiable cases, is comparable with other series
[11, 12]. The simple user interface allowing the
point and click is very easy to use on a practical
319

M Brown, C Eccles and M G Wallis

Figure 4. Distribution of false negative interval cancers for selected readers.

level. It also has only a minimal demand on

memory space when compared with describing the
position of cancer as either full digitized images or
very compressed images. The single point allows
for combining data to identify patterns but is
admitted to be a crude method for marking a
320

three-dimensional mass. This is only a practical

problem when locating a diffuse process such as
the microcalcification of ductal carcinoma in situ.
It provides a greater degree of flexibility than
Hackshaws method of using a grid to divide the
breast disc into eight sectors [10].
The British Journal of Radiology, April 2001

Interval cancer geography: a database

Table 2. Statistical significance of distributions of cancers
Distributions for comparison

View

Significantly
different?

Level of
significance

False negative and true interval cancers

Craniocaudal
Oblique
Craniocaudal
Oblique
Craniocaudal
Oblique

No
No
No
Yes
Yes
Yes

1%
5%
1%

False negative interval cancers and screen

detected cancers
True interval cancers and screen detected cancers

Hanley [13] originally described the distribution

of 1000 breast cancers within the breast
(Figure 5), indicating that 38.2% were situated
in the upper outer quadrant. Lundgren and
Jackobsson [14] initially described screening with
the single mediolateral oblique mammogram,

Figure 5. Hanleys distribution of cancers within the

breast.

which was used as the basis of the UK

NHSBSP until 1995 when the craniocaudal view
was introduced for prevalent screen, as a number
of papers indicate its importance for the additional detection of small cancers [15, 16]. The way
individual areas of the breast are demonstrated on
the standard mediolateral oblique projection and
craniocaudal projections has been documented by
Lee et al [17], and the distribution of our cases
conforms well with Hanleys description [13].
Tabar (personal communication) describes a
systematic schema for viewing mammograms and
graphically indicates the review areas that require
special attention (Figure 6); namely the milky
way and the retroareolar space on the oblique
projection, and the medial portion of the breast
and no mans land at the back of the breast
behind the glandular tissue on the craniocaudal
projection. Our data reinforce this view as most of
our screen detected and interval cancers fall in
similar areas. Hackshaw et al [10] shows similar
clustering to our own data but with a larger
central distribution. Naylor et al [18] used this
distribution pattern on the craniocaudal view to
advise a neutral radiographic position rather than
either medial or lateral rotation, and our data
would tend to reinforce this as cancers are seen in
both the axillary tail and medial aspects.
Interestingly, Daly et al [9] locate their rescreen

Figure 6.
Tabars
forbidden
zones i.e. areas that require special
attention. a, Milky way 34 cm
wide parallel to edge of pectoral
muscle; b, no mans land, retroglandular space; c, medial half of
the breast on the craniocaudal projection; d, retroglandular area.
The British Journal of Radiology, April 2001

321

M Brown, C Eccles and M G Wallis

cancers that were visible 3 years earlier in very

similar positions.
The number of cases is small and not amenable
to statistical analysis when looking at the results
for individual readers (Figure 4). However, as
with the distribution of true, false negative and
control groups, they appear to be distributed in
very similar patterns. So, to date, no obvious
blind spots have been identified for any of our
film readers.
In conclusion, we have developed a database
that allows systematic recording of pathological
and radiological information regarding breast
cancers. Additionally, it can record the geographic position of the cancer with minimal
memory requirements. Statistical differences in
the distribution of false negative and screen
detected cancers have been demonstrated and
the stylized diagrams reinforce the importance of
the conventional review areas. Although this has
not identified any blind spots in our own
readers, it nevertheless provides film readers with
a tool to audit their own work. This database is
available at www.bcsicd.fsworld.co.uk

Acknowledgment
Chris Eccles developed the database as part of
his MSc at Coventry University.

References
1. Forrest P. Breast cancer screening: report to the
Health Ministers of England, Wales, Scotland &
Northern Ireland. London: HMSO, 1986.
2. Day N, McCann J, Camilleri-Ferrante C, Britton P,
Hurst G, Cush S, et al. Monitoring interval cancers
in the Breast Screening Programme: The
East Anglian experience. Quality Assurance
Management Group of East Anglian Breast
Screening Programme. J Med Screen 1995;2:1805.
3. Guidelines on the collection and use of breast
cancer data. Sheffield: NHSBSP, publication
number 26, 1993.

322

4. Quality assurance guidelines for radiologists.

Sheffield: NHSBSP, Publication Number 15, 1997.
5. Duncan KA, Needham G, Gilbert FJ, Deans HE.
Incident round cancers: what can we learn? Clin
Radiol 1998;53:2932.
6. Ikeda DM, Andersson I, Wattsgard C, Janzon L,
Linell F. Interval breast cancers in the malmo
mammographic screening trial: radiographic
appearance and prognostic considerations. AJR
1992;159:28794.
7. Martin JE, Moskowitz M, Milbrath JR. Breast
cancers missed by mammography. AJR 1979;132:
7379.
8. Tabar L, Dean PB. Teaching atlas of mammography. New York, NY: Thieme Inc, 1985:514.
9. Daly CA, Apthrop L, Field S. Second round
cancers: how many were visible on the first round
of the UK National Breast Screening Programme,
three years earlier? Clin Radiol 1998;53:258.
10. Hackshaw AK, Wald NJ, Michell MJ, Field S,
Wilson ARM. An investigation into why two-view
mammography is better than one-view in breast
cancer screening. Clin Radiol 2000;55:4548.
11. Simpson W, Neilson F, Young JR. The identification of false negatives in a population of interval
cancers: a method for audit of screening mammography. Breast 1995;4:1838.
12. Sylvester PA, Kutt E, Baird A, Vipond MN, Webb
AJ, Farndon JR. Rate and classification of interval
cancers in the breast screening programme. Ann R
Coll Surg Engl 1997;79:2767.
13. Hanley RS. Carcinoma of the breast. Ann R Coll
Surg Engl 1975;57:5966.
14. Lundgren B, Jackobsson S. Single view mammography, a simple and efficient approach to breast
cancer screening. Cancer 1976;38:11249.
15. Wald NJ, Murphy P, Major P, Parkes C, Townsend
J, Frost C. UKCCCR multicentre randomised
control trial of one and two view mammography
in breast cancer screening. BMJ 1995;311:118993.
16. Blanks RG, Moss SM, Wallis MG. A comparison
of two view and one view mammography in the
detection of small invasive cancers: results from the
NHSBSP. J Med Screen 1996;3:2003.
17. Lee L, Stickland V, Wilson ARM, Roebuck E.
Fundamentals of mammography. London: WB
Saunders, 1995:2530.
18. Naylor SM, Lee L, Evans AJ. A study to find the
optimal orientation of the breast for the craniocaudal view, for screening purposes. Clin Radiol
1999;54:8046.

The British Journal of Radiology, April 2001